fix(moonshot): add reasoning_content field support for Kimi models (#11806)

Kimi-K2.5 returns 400 errors when thinking is enabled because
reasoning_content is missing from assistant tool call messages.
This adds the supportsReasoningContentField flag (as a getter gated
on kimi-* models) matching the existing DeepSeek fix, so legacy
moonshot-v1-* models are unaffected.

Fixes #11600
Fixes #11711