fix(ai): fit OpenRouter reasoning budgets to the output limit - #51189

Issue for this PR

N/A. Found while testing default output limits for #51021.

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

OpenRouter now fits reasoning.max_tokens to half the output limit with ProviderShared.fitThinkingBudget, the same rule the other protocols got in #51157.

OpenRouter forwards reasoning.max_tokens as the upstream thinking budget, but its body builder copied the reasoning option through unchanged. Upstreams such as Alibaba and Anthropic reject a budget that is not below the output limit. OpenRouter's Qwen max variant sends a 131,071 budget, so a 32k compaction summary failed:

max_completion_tokens [32000] must be greater than thinking_budget [131071]

• Minimum: 1,024, Anthropic's minimum, which also covers Gemini's 512.
• Passes through unchanged: effort-only reasoning, and requests without an output limit.

How did you verify your code works?

• Unit test: covers halving, a budget already within half, the 1,024 minimum, no output limit, and effort-only reasoning. The OpenRouter suite passes.
• Live, with feat(core): fit output limits to the context window #51021 applied, on openrouter/qwen/qwen3.8-flash max:
  • Main steps: max_completion_tokens: 131072, reasoning.max_tokens: 65536.
  • Compaction: max_completion_tokens: 32000, reasoning.max_tokens: 16000. The summary completed, where it was rejected before.

Screenshots / recordings

N/A

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR