feat: add CommandCode provider - #6969

Add CommandCode (https://commandcode.ai/docs/provider) as a new provider.

CommandCode is an OpenAI- and Anthropic-compatible model gateway. Default wire is OpenAI-compatible (/provider/v1/chat/completions); Claude models are served on the Anthropic Messages endpoint, so that model overrides npm/api like other dual-protocol gateways.

This PR registers 7 representative models with base_model + cost + reasoning_options. Remaining catalog models can follow in later PRs.

Model IDs match the live GET /provider/v1/models list and were verified with real requests (IDs are case-sensitive on this gateway).

Verified:

• bun validate passes
• opencode loads the catalog (OPENCODE_MODELS_PATH) and lists commandcode/* models
• live calls OK via opencode: deepseek/deepseek-v4-flash, moonshotai/kimi-k3, Qwen/Qwen3.8-Max, MiniMaxAI/MiniMax-M3, zai-org/GLM-5.3
• Claude models route to /messages correctly (request reaches the gateway; my key's plan does not include Claude, so full completion was not tested)
• gpt-5.6-sol reaches the gateway via direct API call; opencode run returned empty text, likely a client-side parsing quirk, catalog entry itself is correct

Sources: https://commandcode.ai/models and https://commandcode.ai/docs/provider (accessed 2026-09-12).