Openai compatibility: Fetch and use from /models endpoint (and some other features)

[tnglemongrass]

• User might have OPENAI_API_BASE and OPENAI_API_KEY given in their envs. Those should be used.
• and /models (or however you can get to that list) should show models from OPENAI_API_BASE/models and the chat should use OPENAI_API_BASE/chat/completions.
• It would be quite nice to give those via cli args as well, as a starting point e.g. opencode --api-base https://api.openai.com/v1 --api-key your_api_key --model gpt-4.1

e..g OPENAI_API_BASE=https://api.openai.com/v1 or OPENAI_API_BASE=https://api.openai.com/v1/

This would also be quite useful to connect against local providers like ollama or koboldcpp or LMStudio.