Databento-dbn

Bug fixes
- Fixed an import error in the Python type stub file

Enhancements
- Improved `Debug` implementation for all record types
- Prices are formatted as decimals
- Fixed-length strings are formatted as strings
- Bit flag fields are formatted as binary
- Several fields are formatted as enums instead of their raw representations
- Improved `Debug` implementation for `RecordRef` to show `RecordHeader`
- Added `--schema` option to `dbn` CLI tool to filter a DBN to a particular schema. This
allows outputting saved live data to CSV
- Allowed passing `--limit` option to `dbn` CLI tool with `--metadata` flag
- Improved performance of decoding uncompressed DBN fragments with the `dbn` CLI tool
- Added builders to `CsvEncoder`, `DynEncoder`, and `JsonEncoder` to assist with the
growing number of customizations
- Added option to write CSV header as part of creating `CsvEncoder` to make it harder
to forget
- Added `-s`/`--map-symbols` flag to CLI to create a `symbol` field in the output with
the text symbol mapped from the instrument ID
- Added `version` param to Python `Metadata` constructor choose between DBNv1 and DBNv2
- Implemented `EncodeRecordTextExt` for `DynEncoder`
- Implemented `Deserialize` and `Serialize` for all records and enums (with `serde`
feature enabled). This allows serializing records with additional encodings not
supported by the DBN crate
- Implemented `Hash` for all record types
- Added new publisher value for OPRA MIAX Sapphire
- Added Python type definition for `Metadata.__init__`
- Added `metadata_mut` method to decoders to get a mutable reference to the decoded
metadata
- Improved panic message on `RecordRef::get` when length doesn't match expected to be
actionable
- Added `encode::ZSTD_COMPRESSION_LEVEL` constant

Breaking changes
- Increased size of `SystemMsg` and `ErrorMsg` to provide better messages from Live
gateway
- Increased length of `err` and `msg` fields for more detailed messages
- Added `is_last` field to `ErrorMsg` to indicate the last error in a chain
- Added `code` field to `SystemMsg` and `ErrorMsg`, although currently unused
- Added new `is_last` parameter to `ErrorMsg::new`
- Decoding these is backwards-compatible and records with longer messages won't be
sent during the DBN version 2 migration period
- Renamed previous records to `compat::ErrorMsgV1` and `compat::SystemMsgV1`
- Split `DecodeDbn` trait into `DecodeRecord` and `DbnMetadata` traits for more
flexibility. `DecodeDbn` continues to exist as a trait alias
- Moved `decode_stream` out of `DecodeDbn` to its own separate trait `DecodeStream`
- Changed trait bounds of `EncodeDbn::encode_decoded` and `encode_decoded_with_limit` to
`DecodeRecordRef + DbnMetadata`

Bug fixes
- Fixed panic in `TsSymbolMap` when `start_date` == `end_date`
- Added missing Python `__eq__` and `__ne__` implementations for `BidAskPair`
- Fixed Python `size_hint` return value for `InstrumentDefMsgV1` and
`SymbolMappingMsgV1`
- Fixed cases where `dbn` CLI tool would write a broken pipe error to standard error
such as when piping to `head`
- Fixed bug in sync and async `MetadataEncoder`s where `version` was used to determine
the encoded length of fixed-length symbols instead of the `symbol_cstr_len` field

Enhancements
- Added `set_upgrade_policy` setters to `DbnDecoder`, `DbnRecordDecoder`,
`AsyncDbnDecoder`, and `AsyncDbnRecordDecoder`
- Added `from_schema` classmethod for Python `RType` enum

Breaking changes
- Renamed parameter for Python Enum classmethod constructors to `value` from `data`.

Enhancements
- Added new trait `compat::SymbolMappingRec` for code reuse when working with
both versions of `SymbolMappingMsg`
- Changed `PitSymbolMap::on_symbol_mapping` to accept either version of
`SymbolMappingMsg`

Bug fixes
- Fixed missing DBNv1 compatibility in `PitSymbolMap::on_record`
- Fixed missing Python export for `VersionUpgradePolicy`
- Fixed missing Python export and methods for `InstrumentDefMsgV1` and
`SymbolMappingMsgV1`
- Fixed bug where Python `DbnDecoder` and `Transcoder` would throw exceptions
when attempting to decode partial metadata

Enhancements
- This version begins the transition to DBN version 2 (DBNv2). In this version, the
decoders support decoding both versions of DBN and the DBN encoders default to
keeping version of the input. However, in a future version, decoders will by default
convert DBNv1 to DBNv2 and support will be dropped for encoding DBNv1.
- Affects `SymbolMappingMsg`, `InstrumentDefMsg`, and `Metadata`. All other record
types and market data schemas are unchanged
- Version 1 structs can be converted to version 2 structs with the `From` trait
- Added `symbol_cstr_len` field to `Metadata` to indicate the length of fixed symbol
strings
- Added `stype_in` and `stype_out` fields to `SymbolMappingMsg` to provide more context
with live symbology updates
- Added smart wrapping to `dbn` CLI help output
- Updated `rtype_dispatch` family of macros to check record length to handle both
versions of records. This is temporary during the transition period
- Added `VersionUpgradePolicy` enum and associated methods to the decoders to
allow specifying how to handle decoding records from prior DBN versions
- Added `Metadata::upgrade()` method to update `Metadata` from a prior DBN version to
the latest version
- Added `-u`/`--upgrade` flags to `dbn` CLI that when passed upgrades DBN data from
previous versions. By default data is decoded as-is
- Made `AsyncDbnDecoder::decode_record`, `AsyncDbnDecoder::decode_record_ref`,
`dbn::AsyncRecordDecoder::decode`, and `dbn::AsyncRecordDecoder::decode_ref`
cancellation safe. This makes them safe to use within a
`tokio::select!`(https://docs.rs/tokio/latest/tokio/macro.select.html) statement
- Added documentation around cancellation safety for async APIs
- Improved error messages for conversion errors
- Added `TOB` flag to denote top-of-book messages
- Added new publisher values in preparation for IFEU.IMPACT and NDEX.IMPACT datasets
- Added new publisher values for consolidated DBEQ.BASIC and DBEQ.PLUS
- Added `MAX_RECORD_LEN` constant for the length of the largest record type
- Exposed record flag constants in `databento_dbn` with `F_` prefix
- Added export to Python for `RType`

Breaking changes
- The old `InstrumentDefMsg` is now `compat::InstrumentDefMsgV1`
- `compat::InstrumentDefMsgV2` is now an alias for `InstrumentDefMsg`
- The old `SymbolMappingMsg` is now `compat::SymbolMappingMsgV1`
- `compat::SymbolMappingMsgV2` is now an alias for `SymbolMappingMsg`
- Changed `SYMBOL_CSTR_LEN` constant to 71. Previous value is now in
`compat::SYMBOL_CSTR_V1`
- Changed `DBN_VERSION` constant to 2
- `security_update_action` was converted to a raw `c_char` to safely support adding
variants in the future
- Renamed `_dummy` in `InstrumentDefMsg` to `_reserved`
- Removed `_reserved2`, `_reserved3`, and `_reserved5` from `InstrumentDefMsg`
- Removed `_dummy` from `SymbolMappingMsg`
- Moved position of `strike_price` within `InstrumentDefMsg` but left text serialization
order unchanged
- Made `Error` non-exhaustive, meaning it can no longer be exhaustively matched against.
This allows adding additional error variants in the future without a breaking change
- Added `upgrade_policy` parameter to `RecordDecoder::with_version` constructor to
control whether records of previous versions will be upgraded
- Added `upgrade_policy` parameter to `DynDecoder` constructors to control whether
records of previous versions will be upgraded
- Renamed `symbol_map` parameter for Python Transcoder to `symbol_interval_map` to
better reflect the date intervals it contains

Deprecations
- Deprecated unused `write_dbn_file` function from Python interface. Please use
`Transcoder` instead

Bug fixes
- Fixed typo in Python type definition for `InstrumentDefMsg.pretty_high_limit_price`
- Fixed type signature for `Metadata.stype_in` and `Metadata.stype_out` Python methods
- Fixed incorrect version in `pyproject.toml`

Enhancements
- Added `SymbolMappingMsgV2::new` method
- Added `Record` trait for all types beginning with a `RecordHeader`
- Added new `index_ts` and `raw_index_ts` methods to `Record` trait, which returns the
primary timestamp for a record
- Added `RecordMut` trait for accessing a mutable reference to a `RecordHeader`
- Implemented `PartialOrd` for all record types, based on `raw_index_ts`
- Loosened `DbnEncodable` from requiring `HasRType` to only requiring `Record`. This means
`RecordRef`s and concrete records can be encoded with the same methods

Breaking changes
- Split part of `HasRType` into new `Record` and `RecordMut` traits, which are object-
safe: they can be used in `Box<dyn>`. `RecordRef` also implements `Record`, so it's
easier to write code that works for both concrete records as well as `RecordRef`
- Removed `RecordRef` methods made redundant by it implementing `Record`
- Removed `input_compression` parameter from Python `Transcoder`

Deprecations
- Deprecated `SymbolIndex::get_for_rec_ref`, which was made redundant by loosening the
trait bound on `SymbolIndex::get_for_rec` to accept `RecordRef`s

Bug fixes
- Fixed `TsSymbolMap` not always using the correct timestamp for getting the mapped
symbol