feat(stt): make OpenAI transcription model configurable

[xianjianlf2]

Summary

• add [openai].stt_model, defaulting to whisper-1 for backward compatibility
• pass the configured model through the OpenAI STT client into each transcription request
• add config default/override tests and an HTTP-level multipart request regression test
• document the setting and supported transcription models in English and Chinese config/provider docs and READMEs

Closes #25

Verification

• gofmt -l . (no output)
• go build ./...
• go vet ./...
• go test -race ./...
• git diff --check

The request test captures the actual multipart payload sent through go-openai and verifies model=gpt-4o-transcribe; no live API key or network provider was used.

[jason-shen]

@xianjianlf2 thank you!