feat(foundation): implement Gemini streaming support

[Bhanudahiyaa]

Summary

Implements SSE streaming support for the Gemini LLM provider (GeminiProvider::chat_stream) in the mofa-foundation crate. #1272

Motivation

This aligns with GSoC 2026 open task #30 (Streaming response optimization) and Idea #3 (Cognitive Compute Mesh/Multi-provider support). Previously, the Gemini
chat_stream
function returned an explicit ProviderNotSupported error. It now uses reqwest and futures::stream::unfold to parse Gemini SSE streams natively.

Changes

1. Implemented parse_gemini_sse to parse newline-delimited JSON alt=sse structures.
2. Extracted shared request body logic into build_request_body.
3. Updated GeminiProvider::chat_stream to connect to streamGenerateContent?alt=sse.
4. Added 6 synthetic unit tests to cover SSE parsing, configuration builders, and finish reasons.
5. Updated supports_streaming to true and the ModelCapabilities accordingly.

Related Issues

Closes #1272

Testing

• Added unit tests
• Added integration tests
• Tested locally (Passed cargo test -p mofa-foundation, cargo clippy, and cargo fmt)

Checklist

• I have read the CONTRIBUTING.md document.
• I have run cargo fmt --check and cargo clippy.
• I have added tests to cover my changes.
• I have updated the documentation accordingly.

[Bhanudahiyaa]

Hi @BH3GEI @lijingrs
This PR implements SSE streaming support for the Gemini provider in mofa-foundation, enabling GeminiProvider::chat_stream through the streamGenerateContent?alt=sse endpoint.

This contribution aligns with Open Task #30 (Streaming response optimization) and supports the multi-provider inference abstraction described in Idea 3 (Cognitive Compute Mesh).

Please let me know if the streaming abstraction or parser structure should be adjusted to better fit the provider interface. Happy to iterate based on feedback.