fix(dfu): harden flash_beacon.sh with timeouts + retries

[ak2k]

mcumgr was previously called with no timeout flags and no wall-clock
cap. A stuck SMP transfer (BLE link drop, MTU stall, peer reboot
mid-upload) would hang the script indefinitely with no useful output.

Hardening:

• set -euo pipefail (script previously continued past a failed upload
  to image confirm, which would confirm against potentially stale state)
• mcumgr -t 5 -r 5 on every call: 5s per-RPC timeout, 5 retries per
  packet — covers transient MTU/link issues without masking a real hang
• timeout 300s wall-clock cap on the upload step (SIGTERM, then SIGKILL
  after 10s grace) — 200 KB image at MTU 498 is 30-90s in practice
• MCUMGR env var with sane fallback (was hardcoded to ~/go/bin/mcumgr)
• UPLOAD_TIMEOUT env var for tuning
• bash array variables for connection args (eliminates shell-quoting
  fragility around the connstring)
• Argument validation refactored to use ${VAR:-} so unset args fail
  cleanly under set -u
• sudo -v to prime the credential cache (was: sudo echo "")
• shellcheck clean

Behavior on success is unchanged. Failure modes that previously hung
silently now exit non-zero with a clear cause.