fix: Keep sync_container_lifecycles() bgtask alive in a loop.

[fregataa]

Why resource sync API depends on this PR?

TL; DR

sync_containers_lifecycle() should work

Detail

We have 3 different sources of Kernel / Container data

1. Manager side DB
2. Agent side kernel_registry
3. Agent side actual docker containers

I tried to synchronize ALL Agent's data to Manager's DB but synchronizing actual containers to kernel_registry or manager's data needs to process container destruction tasks which take a long time.
Since sync_containers_lifecycle() task synchronizes actual containers to kernel_registry periodically in background, resource-sync API only needs to sync kernel_registry to Manager side DB if it is ensured that the background task stays alive and works as we expect.

Changes

Fix

• Keep sync_container_lifecycles() bgtask alive in a loop by handling any exceptions raised in the task.
• In the try-except context of container creation, do not use unbound local variable

Enhance

• Fetch all containers eagerly in sync_container_lifecycles()

Checklist: (if applicable)

• Milestone metadata specifying the target backport version