[DOC] Clarify architecture diagram — gce/response vs SSE confusion

[OppaAI]

Summary

The architecture diagram in README.md makes it appear there are two separate output paths from CNC to GUI when there is only one.

Location

README.md — Architecture flowchart (Mermaid diagram)

Issue

Diagram shows:

COSMOS -->|SSE tokens| CNC
CNC -->|chunks| gce/response
gce/response --> GUI

This looks like SSE tokens go somewhere separately from gce/response. In reality CNC receives SSE tokens internally from Cosmos and publishes them as JSON chunks to /gce/response — the only output path.

Suggested Fix

Clarify the diagram to show one single output path and add a note explaining the chunk format:

{type: start} → first token
{type: delta} → subsequent tokens
{type: done}  → full response complete
{type: error} → error occurred

Notes

• Flagged by Kimi K2.5 architecture review
• No code changes needed — documentation only