Stream chat responses from Gemini to the browser with SSE

[devangb3]

The chat UI currently waits for the full Gemini response before displaying anything, which for a multi-paragraph answer is 3–8 seconds of spinner. Every production chat app streams. Streaming with SSE is far cheaper than migrating to WebSockets.

Current state:
/chat blocks until Gemini returns, then returns the full answer.

Proposed implementation:

1. Switch Gemini call to streamGenerateContent endpoint.
2. Flask route returns Response(generator, mimetype='text/event-stream') emitting data: {token}\n\n chunks.
3. Frontend uses EventSource (or fetch + ReadableStream, because EventSource is GET-only and the query is in the POST body — easier to POST and parse chunks manually).
4. Source-attribution footer is sent as a final SSE event after the tokens.

Files likely affected:

• app.py (chat route)
• backend/rag_utils.py (LLMIntegration.generate_answer becomes a generator)
• frontend/src/components/ChatInterface.js
• frontend/src/utils/api.js

Acceptance criteria:

• First token visible in the browser within ~500 ms of sending the question.
• Sources and chunks_found still render correctly after the stream completes.
• Aborting the request (user navigates away) closes the upstream Gemini connection.