v0.11.8 - Kimi K2.6 thinking & provider-prefixed model slugs

Highlights

• Kimi K2.6 thinking support via the OpenAI-compatible API. Sends the thinking: { type: enabled|disabled } extension and skips reasoning_effort (which K2.6 does not accept).
• Provider-prefixed model slugs are now recognized for thinking-control routing. Models such as dashscope/deepseek-v4-pro and moonshotai/kimi-k2.6 are correctly matched against their family rules instead of being treated as unknown.

Changes

• krew-llm (openai_chat.rs):
  • Added model_slug / slug_matches_family helpers that strip a provider/ prefix before family matching.
  • is_deepseek_v4 now matches both deepseek-v4-* and dashscope/deepseek-v4-*.
  • New is_kimi_k26 family check covering kimi-k2.6 and moonshotai/kimi-k2.6.
  • Renamed build_deepseek_thinking → build_compatible_thinking, now emitting the thinking extension for both DeepSeek V4 and Kimi K2.6.
  • build_reasoning_effort short-circuits for Kimi K2.6 (no effort field sent).
  • supports_xhigh matches against the slug, so OpenAI xhigh-whitelist models still route correctly when prefixed.

---

Full Changelog: v0.11.7...v0.11.8