feat(ai): add Responses and Gemini support with GPT-5.x and Gemini compatibility improvements

[wgqqqqq]

Summary

• add OpenAI Responses API support across the core client, stream handlers, config layer, and UI model configuration
• add Gemini provider support, including message conversion, streaming response handling, and configuration wiring
• improve GPT-5 / Responses integration with a GPT-5-specific agentic prompt, reasoning effort support, and more reliable streamed tool-call handling
• improve Gemini compatibility by adding native tool support, structured output translation, richer grounding/safety/code-execution metadata handling, and safer fallback behavior when built-in tools would conflict with custom function calling

Why

• enable Responses-based models and Gemini models to work as first-class providers in BitFun instead of relying on OpenAI chat-completions-only behavior
• make GPT-5 and Gemini behave more reliably in real agentic workflows, especially around tool use, streaming, and provider-specific response semantics
• reduce provider-specific failures by normalizing request/response handling while still preserving important native capabilities and metadata

Test plan

• verified targeted Rust tests for Responses and Gemini stream/message conversion paths
• verified Gemini request-building behavior for native-tool-only and mixed-tool scenarios
• verified Gemini parsing for grounding metadata, safety feedback, and code execution response parts
• checked edited files for linter diagnostics