Add APIUsage tracking to extChatEndpoint

[Shinrai]

Token usage reported by third-party language models is currently thrown out entirely — unceremoniously discarded before it can reach the context-window meter. This patch corrects that oversight by plumbing the usage data already present in the LLM response stream through to the appropriate locations within vscode-copilot-chat, so the meter finally gets the numbers it deserves.

---

fix(ext-endpoint): propagate usage from DataPart stream in ExtensionContributedChatEndpoint

The extension-contributed chat endpoint previously returned a hardcoded zeroed-out usage object on every successful response, causing the Copilot Chat context-window meter to never update when routing through third-party language model providers.

Add a usageFromDataPart() helper that checks incoming LanguageModelDataPart chunks for a native usage shape (prompt_tokens, completion_tokens, total_tokens — all numbers) without keying on any MIME type. This mirrors the shape-based check already present in the non-provider (internal) streaming path.

In makeChatRequest2():

• Declare usage: APIUsage | undefined before the stream loop.
• After handling known MIME types (StatefulMarker, ContextManagement), call usageFromDataPart() on every DataPart and capture the result.
• Return usage ?? <zeroed fallback> so existing behaviour is preserved when no usage chunk arrives.

No new MIME branch is introduced; any provider that emits a DataPart whose JSON payload matches the OpenAI usage shape will automatically populate the meter.

---

---

[Shinrai]

@microsoft-github-policy-service agree company="CLDMV"

[Copilot]

Pull request overview

This PR updates the extension-contributed chat endpoint so usage data emitted by third-party language model providers can flow back into Copilot Chat’s normal response pipeline instead of always being replaced with zeros. It fits into the endpoint layer that adapts VS Code LanguageModelChat providers to the extension’s internal ChatResponse/usage model.

Changes:

• Adds a helper in extChatEndpoint.ts to detect OpenAI-shaped usage payloads inside LanguageModelDataPart chunks.
• Captures usage while consuming the provider response stream and returns it on successful responses.
• Keeps the previous zeroed fallback when no usage chunk is present.

[alexdima]

Thanks for the contribution! This repository has been archived because the project has moved into the main VS Code repository.

Could you please reopen/recreate this PR against:
https://github.com/microsoft/vscode/tree/main/extensions/copilot

We’ll continue reviewing contributions there. Thanks!