fix(provider): dedupe streaming usage reports

[Vizards]

Summary

• Dedupe streaming usage emissions in the SSE client and forward only the final usage payload.
• Keep downstream usage reporting simple while avoiding SiliconFlow per-chunk output log spam.
• Guard VS Code usage data reporting so a rejected data part does not block replay marker or cache diagnostics.

Fixes #134

Test plan

• npm run compile
• npm run lint
• npm run format:check

[Copilot]

Pull request overview

This PR improves streaming behavior in the DeepSeek SSE client by deduplicating per-chunk usage emissions (forwarding only the final usage payload) and hardening downstream VS Code usage reporting so failures don’t disrupt replay marker reporting or cache diagnostics.

Changes:

• Accumulate usage during SSE streaming and emit it once at stream completion ([DONE] or EOF).
• Wrap VS Code LanguageModelDataPart usage reporting in a try/catch and log failures with request context.
• Thread requestKind through usage reporting to keep logging consistent with existing request-scoped logs.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/provider/stream.ts	Guards Copilot usage data part reporting so exceptions don’t block replay markers/diagnostics; logs failures with requestKind.
src/client/core.ts	Dedupes streaming usage reports by retaining only the latest usage and emitting it once when the stream completes.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.