Support local coding-agent runners and OpenCode provider

[lesbass]

Which area does this affect?

LLM analysis / insights

What problem are you trying to solve?

Code Insights currently treats native analysis as a Claude Code-only boolean. Users who already authenticate Codex or OpenCode locally cannot reuse those subscriptions for analysis, and must configure a separate API key. The hard-coded OpenAI, Anthropic, and Gemini model suggestions are also outdated.

OpenCode users additionally need first-class access to Zen and Go, both through the authenticated local CLI and through their remote API with a model and API key.

Proposed solution

• Replace the internal native/provider boolean with an extensible runner selector: api, claude, codex, or opencode.
• Keep --native as a backwards-compatible alias for Claude.
• Add isolated, non-interactive Codex and OpenCode runners with structured output.
• Pass runner/model selection through SessionEnd hooks and the analysis queue.
• Add OpenCode Zen/Go as a remote provider with API key, model, base URL, and live model discovery.
• Refresh suggested OpenAI, Anthropic, and Gemini models while continuing to accept arbitrary custom model IDs.

Alternatives considered

Users can configure an API provider separately, but that duplicates authentication and billing when a local coding-agent subscription is already available. Adding only OpenCode as a direct API provider would not solve that use case.

Contribution

• I would be willing to implement this myself and open a PR