Add type-checking guards to DeadlineAlert init - #73294

I had just one thing to point out.

Approving — the guards are a real improvement, and I checked that the reasoning behind the strict callback check holds: _decode_deadline_callback allow-lists AsyncCallback and SyncCallback by qualified class name, so a Callback subclass that passed the old isinstance check only blew up later, at deserialization. Moving that to Dag-parse time is the right call, and the comment recording why isinstance is too loose earns its place.

I'm not gating the merge on the points below, but I would like them fixed — the first two are decisions rather than polish:

• deadline.py:157 — a bare int interval round-trips end to end today (the decoder has a back-compat branch for numbers); this guard rejects it, and the PR carries backport-to-v3-3-test. Normalise it, or drop the backport label and add a newsfragment.
• deadline.py:162 — the comment's own argument applies to the interval check too: pendulum.duration(...) subclasses timedelta, passes the guard, then dies in serde with a TypeError. Normalising subclasses would make it work instead of failing late.
• deadline.py:154 — narrow callback: Callback to AsyncCallback | SyncCallback, so mypy catches the subclass case statically.
• deadline.py:163 — the permitted-callback tuple now duplicates the decoder's allow-list; tie both to one constant.
• test_deadline.py:237 — the interval test never asserts the reported type name, unlike its callback sibling.

Details inline.

Smaller observations

• The PR body says "Adds a couple unit tests…", but the diff also changes two production validation paths in a backport-labelled PR. Worth a line in the description — backport reviewers read it.
• test_serialize_deserialize_deadline_alert: switching the third case to SerializedDeadlineAlert is the right resolution of @SameerMesiah97's point, and gives better coverage of the core-side round trip. The param id serialized_variable_interval now describes a different case though — something like core_serialized_alert would read truer.
• No newsfragment is correct as the PR stands, since the subclass and loose-type cases already failed at (de)serialization. That answer changes if the int rejection above is kept as-is.

This review was drafted by an AI-assisted tool and
confirmed by an Airflow maintainer. The maintainer
approving this PR has read the findings and signed off. If
something feels off, please reply on the PR and a maintainer
will follow up.

More on how Airflow handles maintainer review:
contributing-docs/05_pull_requests.rst.