Fix Dag serialization crash for Task SDK EventsTimetable

[anxkhn]

Any Dag scheduled with the Task SDK EventsTimetable
(from airflow.sdk import EventsTimetable) currently fails to serialize with
SerializationError: Failed to serialize DAG '...': '_summary', so the Dag
cannot be parsed or scheduled.

The SDK timetable serializer in airflow-core/src/airflow/serialization/encoders.py
emits only {event_dates, restrict_to_events, description} and omits the
_summary key, but the core EventsTimetable.deserialize in
airflow-core/src/airflow/timetables/events.py read data["_summary"]
unconditionally and raised KeyError when it was absent. The two are
asymmetric: _summary became a required serialized field in #51203, and the
SDK serializer that omits it was added in #58669.

This makes deserialize retain the summary already computed by
EventsTimetable.__init__ when _summary is absent. Dags using the SDK
timetable serialize and schedule again, while existing serialized data that
still carries _summary keeps its stored value.

A parametrized regression test round-trips a Dag scheduled with the SDK
EventsTimetable through DagSerialization with and without a description
and asserts the summary is preserved.

---

Was generative AI tooling used to co-author this PR?

• Yes

[Vamsi-klu]

Confirmed the root cause: the SDK serialize_timetable for EventsTimetable in serialization/encoders.py emits only event_dates/restrict_to_events/description and omits _summary (which #51203 made a field the core deserialize read unconditionally), so a DAG using from airflow.sdk import EventsTimetable serializes fine but KeyErrors on deserialize -> SerializationError. data.get("_summary", timetable._summary) is safe because core init always sets self._summary (the description if given, else "{n} events"), and existing core-serialized payloads that still carry _summary keep their stored value. The parametrized test reproduces both branches and KeyErrors without the change. One thing worth stating explicitly: the deserialize-side fix is the correct location regardless of any encoder change, because serialized_dag rows already written since #58669 lack _summary, so tolerating its absence on read is required to recover them, whereas emitting _summary only from the SDK encoder would leave already-persisted DAGs broken. FWIW the .get() form is functionally identical to the contextlib.suppress(KeyError) suggestion in the thread. Minor: the newsfragment is named 69283.bugfix.rst but #69283 is an unrelated merged PR, presumably a placeholder to rename to the real number.