fix(desktop): map subagent timeout/error completion to terminal failed state

[DavidMetcalfe]

Summary

Timed-out (or exception-failed) subagents stay stuck as "running" in the desktop Agents panel and status-bar count forever — spinner glyph, counted in "Agents N running", never pruned — even though the parent conversation already received the failure. Fixes the immortal false-active rows reported in #73728 (also addresses #85492, filed as a duplicate; same root cause as #80018 and #81602, which this fix resolves too).

Root cause

Status-vocabulary drift between the Python backend and the desktop store. The backend completes subagents with statuses the UI never learned:

• tools/delegate_tool.py:2479-2488 — hard child timeout (delegation.child_timeout_seconds) emits subagent.complete with status="timeout" (status introduced in fix(delegation): add hard timeout and stale detection for subagent execution #13770)
• tools/delegate_tool.py:2485 — non-timeout exception exit emits the same event with status="error"

The desktop store's asStatus() (apps/desktop/src/store/subagents.ts) was an allowlist of completed | failed | interrupted | queued; anything else fell through to 'running'. A stuck running row is never cleaned up because pruneFinishedSessionSubagents only prunes terminal-status rows at the message.start boundary, and activeSubagentCount keeps counting it in the status bar.

Changes

• apps/desktop/src/store/subagents.ts
  • Map backend "timeout" / "error" completion statuses to the existing terminal failed status; "cancelled" / "canceled" map to interrupted (compatibility aliases).
  • Fail closed on completion: subagent.complete is terminal by definition — an unrecognized status on it now renders failed instead of falling through to running, so a future backend status can't recreate the immortal false-active row. Live events keep the lenient running fallback. (This is the event-aware handling requested in review on fix(desktop): normalise timeout/error subagent statuses to terminal (#73728) #73859.)
  • Synthesize a Timed out after Xs summary (from the backend's duration_seconds) when the raw status is timeout and no summary arrived, so the row explains why it failed instead of rendering as a bare failure.
• apps/desktop/src/store/subagents.test.ts — normalization regression test (timeout/error → failed, cancelled/canceled → interrupted, all prunable), timeout reason synthesis, missing-duration placeholder, event-aware fail-closed vs. lenient live fallback. 14 tests.

Relationship to #73859

This PR cherry-picks the commit from #73859 (by @RelaxJonh — same normalization mapping, authorship preserved in history) and adds the event-aware fail-closed handling that maintainer review requested on #73859 (@teknium1 keep_open, @GottZ triage 08-02/08-03). That PR has had no author response since 07-30; this PR supersedes it so the requested change doesn't stall the fix.

Testing

• npx vitest run src/store/subagents.test.ts — 14 passed (10 pre-existing + 4 new)
• npx eslint src/store/subagents.ts src/store/subagents.test.ts — clean

Notes

Open question: a distinct timeout status with its own glyph/label (instead of mapping to failed) would be richer UI, but is a much larger surface change — type union, glyph mapping, i18n strings, status-bar counts — for the same user-visible outcome. This PR keeps the narrow mapping; the synthesized summary line preserves the reason.

Closes #73728

[alt-glitch]

This was generated by AI during triage.

Duplicate of #73859: both map backend timeout/error completion statuses to a terminal Desktop state; #73859 is the earlier broader implementation with cancelled/canceled coverage.

[DavidMetcalfe]

Superseded by #86592 — @teknium1's salvage built directly on this branch: @RelaxJonh's normalization commit plus the event-aware fail-closed and timeout-summary commits are the first three commits there. Closing to keep the queue unambiguous; coverage and credits are tracked in #86592.