[aksg87]

Description

Summary

• The OpenAI provider sends max_output_tokens as max_tokens. OpenAI
  deprecated that parameter in favor of max_completion_tokens and documents
  it as incompatible with o-series models
  (Chat Completions reference).
  GPT-5 reasoning models reject it as well
  (openai/openai-dotnet#737).
  Callers who set max_output_tokens with those models get a 400 today.
• The provider now sends max_completion_tokens, the documented replacement,
  when requests go to OpenAI's API: no base_url, https://api.openai.com,
  or a regional *.api.openai.com host. It keeps max_tokens for any other
  base_url, because some OpenAI-compatible servers accept only the older
  name. OPENAI_BASE_URL is checked too, since the OpenAI SDK falls back to
  it when base_url is unset. The Batch API path uses the same request
  builder.
• Rewrites the skill doc caveat that told users to omit max_output_tokens
  for o-series models. It still applies to reasoning models behind a custom
  base_url, such as Azure OpenAI deployments.
• Stacked on Route gpt-6 model IDs to the OpenAI provider #599, which rewraps the same skill-doc paragraph. Merge that
  first; this branch then rebases to one commit.

Bug fix

How Has This Been Tested?

Tests

• New: test_max_output_tokens_request_parameter with openai_api,
  custom_base_url, openai_base_url_env, explicit_openai_base_url,
  regional_openai_base_url and lookalike_openai_host cases. The openai_api case fails on main
  (KeyError: 'max_completion_tokens').
• pytest -m 'not live_api and not requires_pip and not integration':
  991 passed, 25 deselected.
• isort 5.13.2 and pyink 24.3.0 checks are clean; pylint is clean on the
  changed files.
• Live: in CI, the only OpenAI live test that sets max_output_tokens is
  test_medication_extraction_with_schema_constraints (512). On this PR it sends max_completion_tokens to
  gpt-4o-mini; on Move OpenAI live tests off retiring GPT-4o models #601 it sends it to gpt-6-luna. The opt-in Batch
  live test is skipped in CI. Pending; added once this run finishes.

Checklist

• I have made any needed documentation changes.
• I have added tests, or I have ensured existing tests cover the changes.
• I have followed Google's Python Style Guide and ran pylint over the
  affected code.