Fix scheduler MySQL task instance index hint

[8silvergun]

closes: #66763

Summary

SchedulerJobRunner._executable_task_instances_to_queued applied USE INDEX (ti_state) twice on MySQL: once on the
inner ranked query that filters by scheduled task instances, and once on the outer rejoin that looks up
task_instance by (dag_id, task_id, run_id, map_index).

The outer query is an exact identity lookup, so forcing ti_state can prevent MySQL from using
task_instance_composite_key. This PR removes only the outer hint and keeps the inner hint that was added for the
original MySQL scheduler optimization.

Regression Coverage

Added a regression test that compiles the scheduler selection query with the MySQL dialect and verifies USE INDEX (ti_state) appears only once.

Tests

• uv run --python 3.12 --group dev pytest airflow-core/tests/unit/jobs/ -xvs
• prek run ruff --from-ref main
• prek run ruff-format --from-ref main

Context

• MySQL Not Using Correct Index for Scheduler Critical Section Query #25627 added the original MySQL ti_state hint for the scheduler state lookup.
• Include the max_active_tasks limit in the query fetching TIs to be queued #54103 reshaped this scheduler query into the ranked subquery plus outer rejoin form; that is where the hint was
  also applied to the outer rejoin.
• Tasks returning from deferred should be prioritized #57210 touched the same scheduling path, but covers deferred-task priority rather than this MySQL hint placement.

---

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

• Yes — OpenAI Codex

Generated-by: OpenAI Codex following [the guidelines](https://github.com/apache/airflow/blob/main/contributing-
docs/05_pull_requests.rst#gen-ai-assisted-contributions)

[eladkal]

cc @xBis7 can you review this change. It was added in #54103

[xBis7]

@eladkal Thanks for pinging me! I'm going to run some tests.

[xBis7]

Generally it looks good. I need to run some benchmark tests just to be sure.