fix(core): preserve prompt cache affinity

[phall1]

Issue for this PR

Closes#42246

Type of change

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

What does this PR do?

V2 generated a stable session cache key but only passed it through the OpenAI provider namespace. This meant OpenRouter did not receive the key its route already supports, while native OpenAI Chat did not lower the key into prompt_cache_key.

This passes the session key to OpenRouter, lowers it for native OpenAI Chat, and leaves generic OpenAI-compatible providers unchanged because they may reject unsupported fields. It restores the cache-affinity behavior described by #39907.

How did you verify your code works?

• 36 focused LLM route tests passed
• 85 SessionRunner tests passed
• bun typecheck passed in packages/llm and packages/core
• full workspace typecheck passed across 30 packages
• Prettier passed for all touched files

Screenshots / recordings

Not applicable; this only changes provider request bodies.

Checklist

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

[github-actions]

Thanks for updating your PR! It now meets our contributing guidelines. 👍