[aksg87]

Description

Summary

• OpenAI shuts down gpt-4o-2024-08-06, gpt-4o-2024-11-20 and
  gpt-4o-mini-2024-07-18 on January 14, 2027. The gpt-4o and
  gpt-4o-mini aliases point at those snapshots. The OpenAI live tests run on
  every push to main and on same-repository PRs, so they would then fail
  with the same 404 that o4-mini returned after its October 23 shutdown
  (fixed in Replace retired o4-mini in the live reasoning-effort test #594).
• Moves DEFAULT_OPENAI_MODEL and the two inline gpt-4o-mini uses (the
  schema-constraints test and the opt-in Batch test) to gpt-6-luna.
• gpt-6-luna defaults to medium reasoning effort
  (model page)
  and accepts temperature only when reasoning_effort is none
  (GPT-6 guide).
  OPENAI_MODEL_PARAMS now sends reasoning_effort: "none" with
  temperature: 0.0, which keeps the runs cheap and close to deterministic,
  and test_explicit_provider_selection passes the same pair. The
  output-schema live test sends neither, so it exercises the model's default
  effort.
• Test-only. It needs Route gpt-6 model IDs to the OpenAI provider #599 (routing) and Send max_output_tokens as max_completion_tokens to OpenAI #600 (max_completion_tokens),
  which have both merged.

Testing

How Has This Been Tested?

Tests

• pytest -m 'not live_api and not requires_pip and not integration':
  993 passed, 25 deselected, the same as main.
• pytest --co tests/test_live_api.py collects all 16 tests.
• isort 5.13.2 and pyink 24.3.0 checks are clean; pylint is clean on the
  changed files.
• With a mocked OpenAI client, the schema-constraints
  request sends model="gpt-6-luna", reasoning_effort="none",
  temperature=0.0, max_completion_tokens=512 and a json_schema response
  format.
• Skipped in CI: test_batch_extraction_uses_openai_batch_api is opt-in
  (LANGEXTRACT_RUN_OPENAI_BATCH_LIVE_TESTS=1) and CI does not set it, so the
  Batch API path on gpt-6-luna is not verified by this PR's live run.
• Live results: 12 passed, 4 skipped (the Batch test above and 3 Vertex AI
  tests, which CI has no credentials for).

Checklist

• I have followed Google's Python Style Guide and ran pylint over the
  affected code.