Critical fix (found in production testing right after 0.20.4-beta): resync-project…

v0.20.5-beta

Critical fix (found in production testing right after 0.20.4-beta): resync-project (triggered after every scale, network change, etc.) could fail indefinitely with No such image if an add-on (e.g. Redis) had never had its image pulled on that particular host. docker compose up -d ran with --pull never: the first scale/resync touching that project after the add-on was added failed every time, and — before the 0.20.4-beta fix that makes this resync mandatory — the failure was silently swallowed, masking the problem. Switched to --pull missing: pulls only the missing image, no impact on the common case (image already cached).