[Feature Request] Implement typing/streaming effect for AI responses in Chat UI

[SyncWithRaj]

🚀 Feature Request

Currently, in the Pipali chat interface, AI responses are displayed entirely all at once after the backend finishes generating them. For longer answers, this creates a perceived delay for the user and feels less interactive compared to standard modern AI chat interfaces.

Describe the solution you'd like
I propose adding a progressive typing or streaming effect for AI messages. The text should appear dynamically, making the app feel significantly faster, more responsive, and natural.

Technical Approach & Questions
Since I want to submit a PR for this, I'd love to align with the current architecture (React 19 + Bun + WebSockets) before writing the code. There are two ways to tackle this:

1. Real-time Streaming: If the backend Director and WebSocket routes already support emitting partial tokens for the ATIF step, I can hook the frontend up to stream the text as it arrives.
2. UI-Simulated Effect: If the backend currently only sends the completed ATIF step at the end of the thought loop, I can create a React component/hook to simulate a fast character-by-character typing effect on the frontend once the payload is received.

Additional context
I am ready to work on this and raise a PR! Could the maintainers let me know which of the two approaches above fits best with the current roadmap?

[SyncWithRaj]

Update: I've successfully implemented this using Approach 1 (Real-time Streaming)! 🚀

After digging into the architecture, I decided to go with native, real-time token streaming rather than a simulated UI effect. It provides a much faster and more authentic experience that perfectly aligns with the generation speed of the LLM.

Here is a high-level overview of the implementation:

• Backend LLM Layer: Tapped into the OpenAI client.responses.stream to listen for response.output_text.delta events inside sendMessageToGpt.
• Agent Piping: Plumbed an onTextChunk callback through the Director, ResearchRunner, and RunExecutor so the deltas can flow up from the model execution seamlessly.
• WebSocket Protocol: Introduced a new text_delta message type to the WebSocket protocol (ServerMessage) and added logic in the ConversationEventBus to publish these chunks in real time.
• Frontend State Machine: Updated useWebSocketChat.ts to natively support a TEXT_DELTA action in the chatReducer. It intelligently locates the active isStreaming assistant message and progressively appends the incoming characters to the content string.

Since the React component (MessageItem.tsx) simply renders message.content via Markdown, the text now smoothly types out natively as tokens arrive over the wire!

I'll be opening a PR with these changes shortly. Let me know if the maintainers have any feedback on this approach!