Pathway

Added
- class `Behavior`, a superclass of all behavior classes.
- class `ExactlyOnceBehavior` indicating we want to create a `CommonBehavior` that results in each window producing exactly one output (shifted in time by an optional `shift` parameter).
- function `exactly_once_behavior` creating an instance of `ExactlyOnceBehavior`.

Changed
- **BREAKING**: `WindowBehavior` is now called `CommonBehavior`, as it can be also used with interval joins.
- **BREAKING**: `window_behavior` is now called `common_behavior`, as it can be also used with interval joins.
- Deprecating parameter `keep_queries` in `pw.io.http.rest_connector`. Now `delete_completed_queries` with an opposite meaning should be used instead. The default is still `delete_completed_queries=True` (equivalent to `keep_queries=False`) but it will soon be required to be set explicitly.

Added
- A flag `with_metadata` for the filesystem-based connectors to attach the source file metadata to the table entries.
- Methods `pw.debug.table_from_list_of_batches` and `pw.debug.table_from_list_of_batches_by_workers` for creating tables with defined data being inserted over time.

Changed
- **BREAKING**: `pw.debug.table_from_pandas` and `pw.debug.table_from_markdown` now will create tables in the streaming mode, instead of static, if given table definition contains `_time` column.
- **BREAKING**: Renamed the parameter `keep_queries` in `pw.io.http.rest_connector` to `delete_queries` with the opposite meaning. It changes the default behavior - it was `keep_queries=False`, now it is `delete_queries=False`.

Added
- A method `get_nearest_items_asof_now` in `KNNIndex` that allows to get nearest neighbors without updating old queries in the future.
- A method `asof_now_join` in `Table` to join rows from left side of the join with right side of the join at their processing time. Past rows from left side are not used when new data appears on the right side.

Added
- `interval_join` now supports forgetting old entries. The configuration can be passed using `behavior` parameter of `interval_join` method.
- Decorator `table_transformer` for marking that functions take Tables as arguments.
- Namespace for all columns `Table.C.*`.
- Output connectors now provide logs about the number of entries written and time taken.
- Filesystem connectors now support reading whole files as rows.
- Command line option for `pathway spawn` to record data and `pathway replay` command to replay data.

Fixed
- `select` operates only on consistent states.

Added
- `Schema` method `typehints` that returns dict of mypy-compatible typehints.
- Support for JSON parsing from CSV sources.
- `restrict` method in `Table` to restrict table universe to the universe of the other table.
- Better support for postgresql types in the output connector.

Changed
- **BREAKING**: renamed `Table` method `dtypes` to `typehints`. It now returns a `dict` of mypy-compatible typehints.
- **BREAKING**: `Schema.__getitem__` returns a data class `ColumnSchema` containing all related information on particular column.
- **BREAKING**: `tuple` reducer used after intervals_over window now sorts values by time.
- **BREAKING**: expressions used in `select`, `filter`, `flatten`, `with_columns`, `with_id`, `with_id_from` have to have the same universe as the table. Earlier it was possible to use an expression from a superset of a table universe. To use expressions from wider universes, one can use `restrict` on the expression source table.
- **BREAKING**: `pw.universes.promise_are_equal(t1, t2)` no longer allows to use references from `t1` and `t2` in a single expression. To change the universe of a table, use `with_universe_of`.
- **BREAKING**: `ix` and `ix_ref` are temporarily broken inside joins (both temporal and ordinary).
- `select`, `filter`, `concat` keep columns as a single stream. The work for other operators is ongoing.

Fixed
- Optional types other than string correctly output to PostgreSQL.