feat(opencode): use adaptive thinking effort for kimi family on anthr…

[chouqin]

Issue for this PR

Closes #

Type of change

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

What does this PR do?

Kimi/Moonshot's Anthropic-compatible endpoints implement Anthropic's adaptive thinking contract (thinking.type="adaptive" + output_config.effort, including xhigh and the display field). Today opencode sends manual thinking
with a hardcoded budgetTokens: 16000 default for kimi models on @ai-sdk/anthropic, and exposes no effort variants for them.

Changes:

• Default options for kimi family models on the anthropic protocol now send {type: "adaptive", display: "summarized"} with effort: "high" instead of budget tokens. display: "summarized" matters because newer adaptive models
  default to omitting thinking text, and that text is what gets replayed on subsequent requests.
• variants() exposes low/medium/high/xhigh/max for kimi models on @ai-sdk/anthropic, each mapping to adaptive thinking + effort.
• anthropicEffort treats kimi family as adaptive. This also fixes models.dev-declared effort options (e.g. kimi-k3 low/high/max) being silently dropped — the anthropic branch previously returned undefined for non-Claude model
  ids.

Why it works: @ai-sdk/anthropic already lowers providerOptions.anthropic.thinking = {type: "adaptive"} and providerOptions.anthropic.effort to thinking + output_config.effort on the wire, so no SDK change is needed. Kimi
family detection matches providerID / api id / model id / api url (isKimiFamily). OpenAI-compatible kimi paths are untouched, and user-selected variants still override the default (merge order unchanged).

Note: an identical isKimiFamily helper also appears in my other PR (preserve unsigned thinking). They are independent; whichever lands second rebases to keep a single copy.

How did you verify your code works?

• Unit tests in test/provider/transform.test.ts: default adaptive+effort for kimi on anthropic npm (asserting no budgetTokens), no thinking default for non-reasoning kimi / non-kimi anthropic / kimi on openai-compatible, the
  five effort variants' shape, and models.dev effort options mapping for a kimi-k3-like entry. bun test test/provider/transform.test.ts → 352 pass, plus bun typecheck from packages/opencode.
• End-to-end through the installed @ai-sdk/anthropic@3.0.82 with a stub fetch capturing the request body: with the providerOptions this PR emits, the outgoing body contains thinking: {"type":"adaptive","display":"summarized"}
  and output_config: {"effort":"high"}. (No live Moonshot call — worth a quick manual check with a real key before merge.)

Screenshots / recordings

If this is a UI change, please include a screenshot or recording.

Checklist

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

If you do not follow this template your PR will be automatically rejected.