feat(providers/openai): allow reasoning-model overrides and recognize gpt-6+ - #57

Why

Bedrock Mantle and direct OpenAI both expose gpt-6-astra, but the fork's Responses client still classifies only gpt-5* (and o-series/codex) as reasoning models. For any newer generation it omits the reasoning parameter entirely (effort and summary are silently dropped) and forwards temperature, which OpenAI rejects with HTTP 400. Upstream charmbracelet#354 (cherry-picked here as-is) only widened the public IsResponsesModel/IsResponsesReasoningModel helpers; getResponsesModelConfig, which gates the request parameters, was left on gpt-5.

Every new model generation currently needs a fork release before chatd can use it correctly. Callers need a way to decide the classification themselves.

What

• Cherry-pick upstream cfb0530 (fix: gpt series 6+ (#354)) unchanged.
• getResponsesModelConfig treats gpt-5 and every later generation (gpt-6, gpt-10, ...) as reasoning models; -chat variants stay non-reasoning; gpt-4 is unchanged.
• New provider option openai.WithReasoningModelFunc(fn func(modelID string) bool), mirroring WithResponsesAPIFunc: when set it replaces the built-in name heuristics for both the Responses client (reasoning params, temperature/top_p filtering, developer vs system role) and the Chat Completions client (WithLanguageModelReasoningModelFunc, including the logprobs/logit_bias filtering in DefaultPrepareCallFunc).

Validation

• go build ./..., go test ./providers/openai/... -count=1, golangci-lint run ./providers/openai/..., gofumpt clean on touched files.
• New tests: TestPrepareParams_ReasoningModelClassification, TestPrepareParams_ChatReasoningModelOverride (plus upstream's TestIsResponsesModel/TestIsResponsesReasoningModel). Mutation checks confirmed each assertion fails when the corresponding change is removed.
• Live probe through the dev AI gateway with this branch: gpt-6-astra now sends reasoning: {"effort":"low"} without temperature and returns 200; forcing non-reasoning on gpt-5.6-luna via the new option sends temperature and gets OpenAI's 400, showing the override takes effect.

Follow-up: coder/coder will bump its pin and expose the override as an openai_config.reasoning_model model-config flag.

Xum acted on behalf of @ibetitsmike for this PR.