Biglist

- Upgrade dependency `upathlib`, removing formats 'json-z' and 'json-zstd'.

Enhanced

- Enhancement and refinement of type annotations.
- Refine the "generic type" annotations for the classes.

Removed

- Removed `BiglistBase.{new_concurrent_iter, concurrent_iter, concurrent_iter_stat, concurrent_iter_done}`. Please use `new_concurrent_file_iter` and the related methods.
- `ParquetBiglist.iter_batches` was removed because it was simple and unnecessary.

Changed and enhanced

- `BiglistBase.load_data_file` losts parameter `mode`. Now it has only one parameter, which is the file path.
- Enhanced `FileView` and related code. Both `ParquetFileData` and `BiglistFileData` are now subclasses of `FileView`.
- Enhanced the treatment of `{ParquetFileData, ParquetBatchData}.scalar_as_py`.
- Class renamings: `FileView` -> `FileReader`; `ParquetFileData` -> `ParquetFileReader`; `BiglistFileData` -> `BiglistFileReader`.
The old names will be available for a period of deprecation.
- Method renamings: `file_view` -> `file_reader`; `file_views` -> `file_readers`.
The old names will be available for a period of deprecation.

Added

- New public method (property) `datafiles_info`.
- Created Sphinx documentation and started hosting it on "Read the Docs". This is the first version to have such documentation. This is the main work of this release.

Fixed

- Fixed a bug in `ParquetBiglist.get_gcsfs` regarding token expiration.

Added

- Added "external" Biglist variant---class `ParquetBiglist`---backed by Parquet data. This exposes a few new classes as part of the public API. To this purpose, there was a code re-org.
- New class `Chain`.
- New class `BiglistFileData`. Although currently the data file content of a `Biglist` is always a Python `list`, the class prepares for possible changes in the future. In addition, this achieves consistency between `Biglist`/`BiglistFileData` and `ParquetBiglist`/`ParquetFileData`.

Removed

- Methods `Biglist.pre_serialize` and `Biglist.post_deserialize` are removed. Alternatives to achieve similar effects are documented. However, the kind of customization facilitated by `pre_serialize` and `post_deserialize` is *discouraged*. The recommendation is to persist in Python built-in types (such as `dict`, `tuple`, etc), and do conversion to/from custom types in application code.
- Removed method `Biglist.destroy`. Persisted data of a temporary `Biglist` (i.e. you did not specify its path when calling `new`) is deleted upon the instance's garbage collection. If you want to delete the storage of a non-temporary `Biglist`, call `rmrf` on its `.path` attribute.

Changed

- More systematic and thoughtful use of `ListView`.
- `Biglist.load_data_file` returns an object of the new class `BiglistFileData`.
- `Biglist.DEFAULT_STORAGE_FORMAT` changed from `pickle` to `pickle-zstd`.
- Reformatted `CHANGELOG`.

- Add `multiplexer` methods.
- Add parameter `require_exists` to `__init__`.
- Add parameter `thread_pool_executor` to `__init__`, hence allowing user to control the number of threads created by this class. This is useful when a large number of `Biglist` instances are active at the same time.

- Removed the "data_info_file" which contains names and lengths of the data files,
as well as indicating their order. Added a new file containing just the number
of data files.

Changed the file name pattern to contain timestamp and length in name,
so that the info previously stored in and loaded from the
"data_info_file" are now parsed out by on-demand iteration over available files.
This change is meant to improve speed during "concurrent writing" to a biglist
with a large number of data files.
- Removed methods `concurrent_append` and `concurrent_extend`. Use `append` and
`extend` instead.
- Made `Biglist` a "generic" class, that is, can take a type parameter in annotations,
like `MyBiglist[MyClass]`.