feat(llm/openai): implement list_models_remote for OpenAI and CompatibleProvider

[bug-ops]

Parent Epic

#991

Depends On

#992

Summary

Override list_models_remote in OpenAiProvider (and by delegation in CompatibleProvider)
to query OpenAI-compatible /v1/models endpoints. This covers OpenAI, Gemini (OpenAI-compat
endpoint), Grok (xAI), Groq, vLLM, and any other CompatibleProvider instance.

Endpoint

GET {base_url}/v1/models
Authorization: Bearer {api_key}

Response Shape (OpenAI standard)

{
  "data": [
    {
      "id": "gpt-4o",
      "object": "model",
      "created": 1715367049,
      "owned_by": "openai"
    }
  ],
  "object": "list"
}

Provider-Specific Notes

Provider	base_url	Notes
OpenAI	https://api.openai.com	Standard; filter to gpt-*, o*, text-embedding-*
Gemini	https://generativelanguage.googleapis.com/v1beta/openai	Returns gemini-* models
Grok (xAI)	https://api.x.ai	Returns grok-* models
Groq	https://api.groq.com/openai	Returns all Groq-hosted models
vLLM / local	configurable	Returns loaded models only

Implementation Notes

• OpenAiProvider::list_models_remote hits {self.base_url}/v1/models.
• Cache slug = sanitized base_url hostname (e.g. "openai_api_openai_com").
• CompatibleProvider::list_models_remote delegates to self.inner.list_models_remote().
• Convert created (unix timestamp) to chrono::DateTime<Utc> for ModelInfo::created_at.
• Do not filter by model type for CompatibleProvider — downstream config can filter.

Acceptance Criteria

• OpenAiProvider::list_models_remote returns full model list from /v1/models.
• CompatibleProvider delegates correctly.
• Cache slug is derived from base_url to avoid collisions between providers.
• Network failure returns LlmError::Network, does not overwrite valid cache.
• Unit tests for slug derivation and response deserialization.