Remove dead AIP-44 trigger-over-BaseSerialization path (DAT.BASE_TRIGGER)

[potiuk]

What

Removes the DAT.BASE_TRIGGER serialization path — encoding/decoding a BaseTrigger instance through BaseSerialization — which is dead code left over from AIP-44's Internal API:

• the encode branch (isinstance(var, BaseTrigger)), the decode branch (type_ == DAT.BASE_TRIGGER), the DAT.BASE_TRIGGER enum value, and the now-unused BaseTrigger import;
• TaskDeferred is also dropped from the exception-encode tuple — TaskDeferred(BaseException)'s only role there was carrying a trigger over the AIP-44 RPC.

Generic AirflowException / BASE_EXC_SER serialization is kept — it has live test coverage and an exception can still reach the serializer via XCom / next_kwargs / the ExtendedJSON column.

Why it's dead

The live deferral path never serializes a trigger through BaseSerialization:

• A task that defers builds a structured DeferTask / TIDeferredStatePayload from defer.trigger.serialize() (the trigger's own classpath + kwargs) — see task_runner._defer_task.
• The execution API stores Trigger(classpath=<str>, encrypted_kwargs=<blob>) without instantiating anything (routes/task_instances.py); the triggerer later imports the classpath from the DB row.
• AIP-44's Internal API — the RPC that serialized runtime objects like triggers/exceptions over the wire — has been removed.

DAT.BASE_TRIGGER was referenced only inside serialized_objects.py + enums.py + tests — no external producer or consumer. (BaseTrigger.hash serializes the trigger's kwargs dict, not the trigger, so it never hit this branch.)

Relationship to #67926 / #68511

• Supersedes Restrict BASE_TRIGGER deserialization to trusted classes (validate before import) #67926 — it gated the now-removed DAT.BASE_TRIGGER decode.
• Complements Restrict exception-node deserialization to known classes without importing the stored name #68511, whose _safe_import_for_deserialize gate on the exception decode (import_string(exc_cls_name)) protects a path this PR keeps live.

[potiuk]

OK. @amoghrajesh @ashb -> I believe (and I made a deeper digging and trying to get historical look at it) we can simply remove the "dead code" that is not used any more.