feat(agent): pluggable LLM provider (openai / gemini)

[nghiadaulau]

What

Replaces the verbose agent.ai.base_url (#245) with a friendly agent.ai.provider enum — openai (default) or gemini — so operators pick a short name instead of a long URL.

config.ProviderBaseURL(provider) maps the provider to its OpenAI-compatible chat-completions endpoint internally (both backends are OpenAI-compatible, so only the base URL changes). Wired through detect + analyze (with a per-task agent.ai.analyze.provider override), config.yaml (env AGENT_AI_PROVIDER), and the Helm chart. Empty = openai, so existing deployments are unchanged.

agent:
  ai:
    enable: true
    provider: gemini              # openai | gemini
    model: gemini-2.5-flash-lite
    api_key: ${AGENT_AI_API_KEY}

Why

base_url meant pasting a long endpoint (https://generativelanguage.googleapis.com/v1beta/openai). A provider name is shorter, less error-prone and self-documenting; adding a backend is one case in ProviderBaseURL.

Supersedes #245

Same goal (bring-your-own LLM / Gemini), cleaner surface — closing #245 in favour of this.

Verified

• go test ./pkg/config/... — provider→endpoint map + per-task override.
• helm lint clean; helm template --set agent.ai.provider=gemini renders provider: "gemini" with no base_url.
• Runtime path is identical to the previously-verified base_url wiring (the OpenAI-compatible URL now comes from ProviderBaseURL), which ran the full detect → Gemini → Telegram loop on minikube.

🤖 Generated with Claude Code