Support providers beyond OpenAI and Anthropic

[obakeng-develops]

meat supports OpenAI and Anthropic. Supporting other providers, or any OpenAI-compatible endpoint, would widen where it can run.

A few cases where this matters:

• Models from other labs such as DeepSeek, Kimi, Qwen or Mistral, which are often cheaper for a job like abridging a diff.
• Corporate accounts on Bedrock, Vertex or Azure OpenAI, where the model is the same but the endpoint and the auth differ.
• Local models through Ollama or llama.cpp, which matters when the diff comes from a private repo and should not leave the machine.
• Gateways such as OpenRouter.

OPENAI_BASE_URL already exists, and DeepSeek and Kimi both publish OpenAI-compatible endpoints, so some of this may work today. If it does, the README saying so would help, because the flag list does not make it obvious. The gaps are auth schemes that are not a bearer token, and providers that are not OpenAI-compatible.

[zshanhui]

the DeepSeek API and other open source models usually use the Chat Completions api format. so another provider needs to be added. i have it setup with Deepseek v4 flash on my local fork and it works pretty great at 1/50th of the cost of GPT-5.6-sol!