web: fix 90s shutdown hang on every deploy

[espadonne]

Every deploy restart produced ~90 s of 502s: Run() logged "shutdown signal received" and then hung until systemd's default TimeoutStopSec SIGKILLed it. The page-cache LISTEN goroutine holds a pool connection and waits on the server's root context, which the web command never cancels, so the deferred pool.Close() blocked forever.

• f4ac673 web: cancel background context on shutdown so pool close cannot hang (also srv.Close() after the drain window so long polls/SSE release)
• a898a2b systemd: bound web stop timeout at 30s
• pagecache: pin that Listen releases its connection on cancel

Test targets:

• CI Integration step: internal/cache/pagecache (new TestListen_ReleasesConnOnCancel)
• After deploy: journalctl -u shithubd-web shows Stopping→Stopped in seconds, not 90 s; 502 count per deploy near zero