fix(desktop): reap orphan gateways at startup

[cadezhou]

Fixes #77276.

🎯Problem

The Desktop backend forks a gateway child on every serve, but only graceful shutdown reaps it. On abnormal exit (crash / SIGKILL / power loss / forced update), the old gateway is reparented to launchd (PPID=1) and keeps holding the QQ WebSocket. The next serve forks a new gateway, and both race the same QQ credential — messages split across parallel session trees.

🎯Fix
_reap_unsupervised_gateway_orphans() (in hermes_cli/gateway.py) already handles exactly this case, but was only wired into the CLI stop path. This PR calls it once at Desktop lifespan startup, before the new gateway is forked — so stale orphans are cleared first.

Desktop-only (guarded by HERMES_DESKTOP=1)
Wrapped in try/except — a reap failure never blocks startup
Lazy import, matching the file's existing style
Reaper itself unchanged (SIGTERM → 5s → SIGKILL, current-profile match, no-op under systemd)
Net ~5 lines — a wiring fix reusing existing code, no new mechanism.

Relationship to #77297
Orthogonal: #77297 covers graceful shutdown, this covers all abnormal-exit paths. Merges independently.

Testing
Added test_desktop_lifespan_reaps_orphan_gateways_on_startup asserting the reaper runs once at startup under HERMES_DESKTOP=1. Existing reaper tests unchanged and green; full dashboard suite (39 tests) passes.

[alt-glitch]

This was generated by AI during triage.

Duplicate of #78002 — both add the same Desktop lifespan startup call to _reap_unsupervised_gateway_orphans() for #77276.

[leonphull]

Same lifecycle idea, serve-shaped cousin

This PR reaps orphan gateways at startup — we hit the same pattern for orphan Desktop-local hermes serve backends (127.0.0.1 + --port 0, ppid 1), which then exhaust nofile and empty the Desktop sidebar/tabs.

Related:

• Incident: Desktop: orphan local hermes serve backends pile up → EMFILE → empty tabs/sidebar #78872
• Serve-side startup reap + parent-death + nofile floor: fix(runtime,desktop,gateway): EMFILE hardening — nofile floor, orphan serve reap, launchd plist limit persistence #78873
• Parent-death watchdog PR: fix(desktop): reap orphaned serve backends via parent-death watchdog + group-kill #73066

complementary, not competing — gateways and serves both need the “reap already-orphaned + don’t leave new orphans” pair.

[teknium1]

Merged via PR #83406 — your commits were cherry-picked onto current main with your authorship preserved in git log (rebase merge). Thank you for the contribution!

The omnibus combined the WAL read-connection pool (#76700), the configurable nofile floor (#77587), the parent-death watchdog (#73066), the Desktop-boot orphan serve reap incl. the backend.lock.json spare-guard (#78873), and the orphan gateway reap (#78312). All pieces were live-tested end-to-end: real orphaned processes reaped on actual Desktop boot with a lock-owned backend surviving, 151→9 connections under 150 reader threads, and zero fd growth over 800 requests against a live serve.