Fix schedule_tis HA race on try_number/state transitions

[sidshas03]

Related to #57618 and #60330.

This updates DagRun.schedule_tis() to make scheduling more robust under HA scheduler races.

• Guard TI updates by both:
  • TI.id IN (...)
  • TI still being in a schedulable state (NULL or SCHEDULEABLE_STATES)
• Reuse one next_try_number expression so behavior is consistent:
  • UP_FOR_RESCHEDULE keeps the same try_number
  • other schedulable states increment try_number
• Apply the same state guard to EmptyOperator short-circuit updates (SUCCESS) to avoid stale scheduler views overwriting newer state.
• Keep existing debug mismatch logging, but skip mismatch warnings when row state is no longer SCHEDULED (expected in race/no-op cases).

In HA setups, two schedulers can race on the same TI. A stale scheduler view should not be able to:

• increment try_number again
• transition state after another scheduler already moved it to QUEUED / RUNNING / etc.

This change makes those stale updates no-ops.

Tests added

In airflow-core/tests/unit/models/test_dagrun.py:

1. test_schedule_tis_does_not_increment_try_number_if_ti_already_queued_by_other_scheduler
2. test_schedule_tis_empty_operator_does_not_short_circuit_if_ti_already_queued
3. test_schedule_tis_up_for_reschedule_does_not_increment_try_number
4. test_schedule_tis_is_noop_if_ti_transitions_to_nonschedulable_state_before_update
5. test_schedule_tis_empty_operator_is_noop_if_ti_already_running
6. test_schedule_tis_only_one_scheduler_update_succeeds_when_competing

Local validation

• Targeted new tests: passed
• Broader -k schedule_tis subset in test_dagrun.py: passed (existing expected xfails unchanged)
• Ruff on modified files: passed