fix(llm): allow disabling prompt cache key

[lihailong00]

Summary

• add a prompt_cache_key boolean setting to kimi provider configuration
• omit the session-derived prompt_cache_key request field when the setting is false
• preserve the current default behavior for managed Kimi providers
• document the setting in English and Chinese configuration guides

Third-party Kimi-compatible endpoints such as NVIDIA NIM can use:

[providers.nvidia]
type = "kimi"
base_url = "https://integrate.api.nvidia.com/v1"
api_key = "..."
prompt_cache_key = false

Reproduction and verification

A local OpenAI-compatible HTTP endpoint reproduces the reported behavior: before the fix it returns HTTP 400 when the request contains prompt_cache_key. With the new setting disabled, the real request JSON completely omits the field. A second HTTP regression verifies that the default managed Kimi path still sends the session ID.

• uv run pytest tests/core/test_config.py tests/core/test_create_llm.py -q — 53 passed
• targeted Ruff — passed
• targeted Pyright — 0 errors
• targeted Ty — passed
• git diff --check — passed

Closes #2534

---

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 1 additional finding.