feat(api): implement GET /v1/models/{model} retrieve model endpoint

[IllusionOfControl]

Description

Summary

Implements the GET /v1/models/{model} endpoint for retrieving individual model metadata, adhering to the OpenAI Models API specification (client.models.retrieve(...)).

Previously, only GET /v1/models (listing all models) and GET /v1beta/models/{model} (Gemini single-model route) were available. Standard OpenAI SDKs and third-party tools querying GET /v1/models/{model} received a 404 Not Found response.

Key Changes

1. Routing (internal/api/router.go):

   • Registered GET /v1/models/{model...} in the public http.ServeMux.
   • Used {model...} wildcard pattern to accommodate both plain model IDs (gemini-2.5-flash) and prefixed/namespaced identifiers (models/gemini-2.5-flash).

2. OpenAI Model Handler (internal/api/openai.go):

   • Extracted openAIModelObject(model aistudio.Model) map[string]any helper to ensure shared schema consistency between model list and retrieve endpoints.
   • Implemented handleOpenAIModel(w http.ResponseWriter, r *http.Request):
     • Resolves models by canonical ID, models/ prefix, and configured aliases (capability_options.aliases).
     • Returns standard OpenAI model object (object: "model", limits, methods, capabilities, paid flags).
     • Returns 404 Not Found with standardized OpenAI error structure (code: "model_not_found", type: "invalid_request_error") if the requested model does not exist.

3. Anthropic Compatibility (internal/api/anthropic.go):

   • Added writeAnthropicModel(w http.ResponseWriter, model aistudio.Model).
   • When requests include the Anthropic-Version header, returns Anthropic-formatted model objects (type: "model"), or not_found_error on 404.

4. Unit Tests (internal/api/models_test.go):

   • Added test suite TestOpenAIModelsEndpoint covering:
     • list models openai: verifying GET /v1/models format.
     • list models anthropic: verifying GET /v1/models with Anthropic-Version.
     • retrieve model by id: verifying GET /v1/models/gemini-2.5-flash.
     • retrieve model with prefix models/: verifying GET /v1/models/models/gemini-2.5-flash.
     • retrieve model by alias: verifying lookup via alias configured in capability_options.aliases.
     • retrieve model not found: validating 404 status and model_not_found error payload.
     • retrieve model anthropic found: validating Anthropic response structure.
     • retrieve model anthropic not found: validating 404 Anthropic not_found_error.

5. Documentation:

   • Updated endpoint reference tables in README.md, README_en.md, docs/protocol.md, and docs/development.md.

Verification

• go test -v ./internal/api -run TestOpenAIModelsEndpoint: all subtests passed.
• go test ./...: all package tests across repository passed.
• go vet ./...: 0 warnings or errors.