feat(provider): add gemini google-generative-ai compatibility

[fuhao009]

Summary

• add provider compatibility groundwork for API mode and custom headers
• add Gemini (google-generative-ai) request/response compatibility
• support Gemini generateContent endpoint and tool-call mapping

Validation

• go build -o ./build/picoclaw ./cmd/picoclaw
• debug run confirms request path: /v1beta/models/{model}:generateContent
• upstream service may return 503 when account pool unavailable

[Leeaandrob]

@Zepan Gemini compatibility provider — useful addition. Same note as other provider PRs: PR #213 (Provider Refactor — roadmap #283) will restructure the provider layer.

Recommendation: Wait for #213 to merge. Gemini could be registered as an OpenAI-compatible provider in the new architecture, making integration simpler.

[yinwm]

Thanks for this contribution! The Gemini native API support is valuable.

However, this PR needs significant rework due to recent architecture changes:

1. Provider Architecture Refactored (feat(config): refactor provider architecture to protocol-based model_list #492)

PR #492 (merged 2026-02-20) introduced a protocol-based model_list configuration. See the design doc:
provider-refactoring.md

2. Current State

The existing gemini protocol uses OpenAI-compatible /chat/completions endpoint:
case "gemini":
return NewHTTPProviderWithMaxTokensField(...) // OpenAI compat

3. Suggested Approach

To add native Gemini API support, please:

1. Create a new protocol prefix (e.g., gemini-native or google-generative-ai)
2. Create pkg/providers/gemini/provider.go following the pattern of anthropic/ or openai_compat/
3. Register the new protocol in pkg/providers/factory_provider.go

Example configuration:

  {
    "model_list": [
      {
        "model_name": "my-gemini",
        "model": "gemini-native/gemini-2.0-flash",
        "api_key": "xxx"
      }
    ]
  }

@fuhao009

[xiaket]

@fuhao009 gentle nudge :)