No code changes: session ended early due quota exhaustion

[Copilot]

This issue targets provider-alignment updates for OpenAI Responses API, Anthropic headers/betas, Gemini generationConfig, and gateway delegation for modern agentic workflows.

No repository changes were committed in this session because execution stopped early due insufficient quota before implementation could proceed.

• Status
  • No code changes
  • No test changes
  • No docs changes

// No diff produced in this session.

Original prompt

Based on official documentation analysis (July 2026):

1. OpenAI: Fully support the Responses API as the primary endpoint for agentic workflows (multi-turn tool calling, streaming deltas for function_call_arguments, built-in tools like web_search/code_interpreter/remote MCP, parallel_tool_calls, strict mode). Ensure temperature is passed correctly and use max_completion_tokens (official recommendation) instead of or in addition to max_tokens. Update buildResponsesRequest and related logic in openai/provider.go and openai_compat layer.

2. Anthropic: Always include the required anthropic-version header. Move Beta features to the anthropic-beta request header (e.g. for code_execution_, web_search_, fine-grained-tool-streaming, managed-agents). Ensure temperature is forwarded. Update header construction in anthropic/provider.go and any internal clients.

3. Gemini: Wrap temperature and maxOutputTokens inside the generationConfig object (required structure). Clean up Bearer API Key handling. Support generationConfig for newer Gemini models (recommend default temperature=1.0 for reasoning models). Update gemini/provider.go accordingly.

4. Gateway (llm/gateway/gateway.go): Ensure unified Invoke/Stream paths correctly delegate to the updated provider adapters for modern agentic patterns (Responses API stateful input, tool streaming events, generationConfig, headers).

Goal: Make the LLM gateway production-ready for current agentic frameworks per official docs from OpenAI, Anthropic, and Google. Maintain backward compatibility where possible. Add or update relevant tests if the changes allow. Create a clear branch and PR with good description referencing the official doc alignments.