fix(chat): don't abort healthy streams during prefill (#6231)

[touzenesmy]

Summary

When a stream has not yet emitted its first token, the visibilitychange tab-recovery handler treats "no bytes received since the tab was hidden" as evidence of a frozen connection — but a slow local / multimodal model can legitimately spend minutes pre-filling (image + large agent prompt) before its first chunk, and the main prefill path sends no SSE heartbeat at all, so lastActivity stays frozen. Returning to the tab after 20s+ therefore aborted a perfectly healthy stream with "interrupted after the tab went inactive".

Fix: track firstOutputAt on the stream entry (set on the first data event) and skip tab-recovery until the stream has produced output:

• no first token yet → normal prefill silence, don't abort;
• had output then froze → recovery works exactly as before.
  A genuinely dead prefill is still caught by the response timeout, not tab-recovery.

Target branch

• This PR targets dev, not main. All PRs land in dev; main is curated by the maintainer at each release. If your PR is on main by accident, click "Edit" on this PR and change the base.

Linked Issue

Fixes #6231 — reported there before filing this PR (no existing report existed: searched open and closed issues for the exact error string, 0 results). Repro steps in How to Test below mirror the issue's Steps to Reproduce.

Type of Change

• Bug fix (non-breaking — fixes a confirmed issue)
• New feature (non-breaking — adds new behaviour)
• Breaking change (changes or removes existing behaviour)
• Refactor / cleanup (behaviour unchanged)
• Documentation only
• CI / tooling / configuration

Checklist

• I searched open issues and open PRs — this is not a duplicate. (Searched the exact error string, tab went inactive, and visibilitychange PRs: perf(ui): pause background polling when tab is hidden #3821 pauses background polling, fix(ui): stop the whirlpool spinner animating when it is never attached #5989/fix(ui): stop the whirlpool spinner animating when it is never attached #5990 spinner — unrelated.)
• This PR targets dev
• My changes are limited to the scope described above — no unrelated refactors or whitespace changes mixed in. (1 file, +11 lines, 3 small hunks.)
• I actually ran the app (docker compose up or uvicorn app:app) and verified the change works end-to-end. Type-checks and unit tests are not enough. (Verified on the running instance: image + large prompt, tab switched away 25s+ mid-prefill, stream survives; frozen-stream regression path unchanged.)
• I did not run the app/runtime validation and stated that gap in How to Test. Leave this unchecked when the app-run box above is checked.

How to Test

1. Repro (without the fix): send a prompt whose prefill takes >20s on a slow/local model — attach an image plus a large prompt. Immediately switch to another browser tab, wait 25s+, return. Before this PR: the stream is killed with "interrupted after the tab went inactive" even though the server is still happily pre-filling. With this PR: the stream continues.
2. Regression (frozen streams still recover): kill the backend mid-stream after the first token, switch tabs 20s+, return. The tab-recovery path fires exactly as before — the firstOutputAt guard only skips streams that never produced output.
3. node --check static/js/chat.js passes. (No unit test — this is frontend stream-state logic; the two behavioural paths above cover it.)

Visual / UI changes — REQUIRED if you touched anything that renders

Anything that changes what the UI looks like — buttons, icons, padding, colors, fonts, spacing, layout, CSS, HTML, SVG, or any static/js/ module that draws to the DOM — needs all of the following. PRs that change rendering without these WILL be closed.

• Screenshot or short clip of the change in the running app, attached below. Mobile screenshot too if the change affects mobile. — Behaviour-only change (stream lifecycle; no DOM/CSS/styling change). Per the template's static/js clause a clip is attached below showing the stream surviving a tab-away during prefill.
• Style match: the change uses Odysseus's existing visual language. Specifically:
  • Reuse existing CSS variables (--red, --fg, --bg, --card, --border, etc.) — do not introduce new color values, font sizes, or spacing units.
  • Reuse existing button/input/card/border classes. Don't invent parallel styling.
  • No Unicode emoji in UI or code. Use inline SVG (matching the monochrome icon style already in static/index.html) or plain text.
  • Monospaced font (Fira Code) for primary UI text. Don't override.
  • Dark theme is the default; any light-mode work must be wired through the existing theme system, not hard-coded.
• No new component patterns. If a similar widget already exists in the app, extend it instead of writing a parallel one.
• I am not an LLM agent submitting a bulk PR. If you are, please open an issue describing the problem first — bulk auto-generated PRs that don't match the project's visual style are closed on sight, even when the underlying fix is correct.

Screenshots / clips

I had a lot of trouble to say below 10Mb

brave_PHihh1TWja.mp4

(clip of the tab-away-during-prefill repro attached before publishing)