Retry autoscaler jobs on terminated instances

What does this MR do?

When the acquired instance is unreachable, the autoscaler executor marks it unhealthy and retries the job on a different instance using taskscaler's new Replace. cleanupNetwork no longer logs Failed to remove network for build when no network was created.

Why was this MR needed?

When an instance went down, a new job could still be assigned to it. Preparing the job on the dead instance failed, and the runner then tried to remove the job's network on that instance, failing with Failed to remove network for build. This happened even with FF_USE_FLEETING_ACQUIRE_HEARTBEATS enabled.

Depends on gitlab-org/fleeting/taskscaler!109 (merged).

What's the best way to test this MR?

go test -race ./executors/internal/autoscaler/ ./executors/docker/

What are the relevant issue numbers?

Closes #39145