Shutdown hangs up to TimeoutStopSec when a poller tick is stuck on slow SNMP #481

Observed on 2.26.1 (install.sh single-binary, systemd):

systemctl stop netpulse hung until systemd SIGKILLed it at 90s (journal: Failed with result 'timeout').

Root cause (code):

• Poller.Stop() waits for the in-flight tick (<-p.doneCh, internal/poller/poller.go:98-103).
• The tick runs with context.Background() and no deadline (internal/poller/poller.go:112); a live tick polling several hosts via SNMP at ~10s per timed-out request (5s timeout + 1 retry, per host/OID serial) can take minutes.
• The 3s lifeline in cmd/netpulse/main.go:553 is armed only AFTER p.Stop()/upd.Stop() return (main.go:534-535), so it never fires while the tick is stuck.

Impact: every restart on a host with a slow/unreachable SNMP device takes 90s+SIGKILL, and anything scripted around restarts (e.g. the stable self-update helper healthcheck) sees much longer service restarts.

Suggested fix: thread a cancellable context through the tick (Stop() cancels it), and/or arm the 3s lifeline BEFORE calling p.Stop()/upd.Stop(). Also consider a per-tick deadline.

Repro: make one SNMP target blackhole (drop packets) so requests time out, then systemctl restart netpulse and watch it hit TimeoutStopSec.