[BUG] Reasoning effort always sent as "medium" (OpenAI-compatible provider)

[DaubnerF]

Problem / Value

When using an OpenAI-compatible provider in the VS Code extension, the reasoning effort chosen in provider settings is never reflected in the actual request — the outgoing request always carries "medium". Any user of the OpenAI-compatible provider who relies on controlling reasoning depth (for cost, latency, or response style) gets a silently ignored setting, with no indication in the extension that their choice had no effect.

Context

Affects users of the VS Code extension (version 3.80.1, latest release at the time) who use the OpenAI-compatible provider with reasoning effort enabled. The user selects a level such as High and saves, but the request sent to the endpoint contains the reasoning effort field explicitly set to "medium", regardless of the configured level. This held across multiple different models used through the same provider configuration, so the behavior does not appear to be specific to one model. (Confirmed here by pointing the provider at a self-hosted routing service whose request logs show the incoming reasoning effort, but the defect is in what the extension sends — it is not specific to any gateway or routing setup.)

Reproduction

1. Open the provider settings in the extension and configure an OpenAI-compatible provider endpoint that can log or echo incoming request bodies (e.g. a local proxy or gateway)
2. Enable the reasoning effort option and select a level (e.g. High), then save
3. Send a message
4. Inspect the logged request
5. Expected result: the request contains the selected reasoning effort level (e.g. High)
6. Actual result: the request contains the reasoning effort field explicitly set to "medium"
7. Variations tried: repeated with multiple different models through the same provider — the request always shows "medium"

Version

3.80.1