Split from #447 (item 4 of 4). incus 7.2 turns a single failing proxy device into a daemon-wide outage: a container stop during startup-autostart deadlocks incusd (all threads in futex_wait on the callhook), systemd SIGKILLs it after waitready(600s)+stop-sigterm(330s), and the cycle repeats forever, leaking a veth pair and a hung 'forknet dhcp' process per ~15.5min iteration (observed on pi-node-003, #445/#447).
Required: (a) upgrade the pi fleet past 7.2 - incus 7.3 contains a related fix (commit 629a88537 'forknet: Wait up to 5s for initial DHCP configuration'); verify the startup-autostart stop-hook deadlock is actually covered, and (b) report the deadlock upstream if not. hydraskin owns incus installation on the nodes, so version pinning/upgrade belongs in its install/update path.