[FEATURE]: Log LLM API request/response body at DEBUG log level #29186

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

Problem

Currently --log-level DEBUG produces no useful output for debugging LLM API calls. Across the entire codebase there are only 21 log.debug() calls (mostly config loading and TUI audio), and none of them cover the LLM request body sent to providers.

When troubleshooting issues like provider errors, malformed requests, or unexpected model behavior, there's no way to inspect what opencode actually sends to the API.

Comparison

Claude Code has --debug "api" which logs API-level request/response details. This is essential for debugging custom providers and understanding model failures.

Proposed Solution

Add log.debug() calls at the session/llm.ts level, after LLMRequestPrep.prepare() returns the prepared request. At minimum, log:

• messages array (system prompt + conversation history)
• tools being sent
• temperature / maxOutputTokens / other params
• response headers/status on error
  Ideally also log at the HTTP layer (e.g. via a custom fetch wrapper in provider.ts) to capture the raw JSON body sent to the provider endpoint.

Affected Files

• packages/opencode/src/session/llm.ts — where streamText() is called
• packages/opencode/src/session/llm/request.ts — where request is assembled
• packages/opencode/src/provider/provider.ts — where provider SDKs are created