Fireworks agent requests fail on unsupported CoT metadata

Type: Bug

**Summary**
VS Code agents using the Fireworks provider fail because internal chain-of-thought metadata is forwarded inside a chat message. Fireworks rejects the request during validation.

**Environment**
Client: VS Code Agents / Copilot SDK
Provider: Fireworks
Model: glm-5p3-flash
Error
Failed to get response from the AI model; retried 5 times
400 invalid_request_error
Extra inputs are not permitted:

messages[2].cot_id
messages[2].cot_summary
The client surfaces the underlying Fireworks 400 as a 502 and retries unnecessarily.

**Steps to reproduce**
Configure a VS Code agent to use the Fireworks model.
Start a multi-turn conversation.
Submit a follow-up request after the agent produces reasoning metadata.
Observe the request failure.
Expected behavior
The client should remove provider-internal metadata before sending messages, or Fireworks should accept/ignore these fields.

**Actual behavior**
Fireworks strictly validates message objects and rejects unknown fields before model execution.

**Impact**
Requests fail after multiple retries.
Users experience approximately 30 seconds of delay.
Disabling reasoning may reduce quality on complex tasks.

VS Code version: Code 1.137.0 (645f29cc3176500b4b5762ba887cf2a7f0ffdf2c, 2026-09-08T13:43:59-07:00)
OS version: Linux x64 6.8.0-139-generic