[Medium] No timeout on SSE/LLM stream — hung connections held indefinitely

[bmersereau]

Problem

backend/src/routes/chat.ts — POST /chat opens an SSE stream and awaits runLLMStream() with no timeout:

res.setHeader("Content-Type", "text/event-stream");
res.flushHeaders();
const { fullText, events } = await runLLMStream({ ... });  // no timeout or AbortSignal

If the upstream LLM API stalls (network partition, provider outage), the SSE connection stays open indefinitely. Under load, a wave of stalled requests can exhaust the process's open-connection limit.

The same applies to POST /projects/:projectId/chat and POST /tabular-review/:reviewId/generate.

Fix

Wrap the LLM call in a timeout using AbortController (e.g., 3 minutes). On timeout:

1. Abort the upstream request.
2. Write an error SSE event ({ type: "error", message: "Request timed out" }).
3. Write data: [DONE] and end the response.