…ainer" enabled

When "Turn Off Container During Backup" is enabled, the service is scaled
to 0 (or the container stopped) before the backup runs. If the backup
step then failed — e.g. Docker could not pull the ubuntu image used for
the tar — the script aborted via `set -e` before the start/scale-up step,
leaving the service offline indefinitely until a manual reload.

Run the backup in a standalone subshell with `set +e` in the parent,
capture its exit status, and always run the start step afterwards. The
job still reports failure (and fires notifications) once the service is
back up, and the S3 upload only runs on success.

The subshell is deliberately NOT the left side of `|| BACKUP_EXIT=$?`:
in that tested context bash disables the subshell's own `set -e`, which
would silently swallow a mid-backup failure and report success. A plain
subshell followed by `$?` keeps `set -e` active inside it.

Applies to both application and compose (stack and docker-compose)
volume backups via a shared guard.

Fixes Dokploy#4263

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>