fix: scope S3/minio cleanup to the current job's backup files

[alwynpan]

Problem

cleanup_old_data() deletes every object older than the retention window under s3://${bucket}/${s3_path}/ with no filename filter:

run_as_user aws ... s3 ls s3://${backup_job_s3_bucket}/${backup_job_s3_path}/ ... | grep " DIR " -v | grep " PRE " -v | while read -r s3_file; do
    ...
    if [[ $s3_createdate -le $s3_olderthan ]] ; then
        s3_filename=$(echo $s3_file | awk {'print $4'})
        ...
        run_as_user aws ... s3 rm s3://${backup_job_s3_bucket}/${backup_job_s3_path}/${s3_filename} ...
    fi
done

When multiple backup jobs (different databases/hosts) share one bucket + path prefix, or unrelated objects live under it, one job's cleanup deletes the other jobs' backups and any unrelated files older than its own cleanup_time — cross-job data loss and incorrect retention.

The file/filesystem and blobxfer branches already scope deletion to the job's own files via -iname "${backup_job_filename_base}*" (and "${backup_job_global_base}*" for postgres globals). The S3/minio branch had no equivalent.

Fix

Apply the same job-scoping to the S3/minio branch: only remove an object when its name matches this job's filename base (case-insensitive prefix), guarded against an empty base so it can't fall back to matching everything. Non-matching objects are logged at debug level instead of deleted.

if { [ -n "${backup_job_filename_base}" ] && [[ "${s3_filename,,}" == "${backup_job_filename_base,,}"* ]] ; } || \
   { var_true "${_postgres_backup_globals}" && [ -n "${backup_job_global_base}" ] && [[ "${s3_filename,,}" == "${backup_job_global_base,,}"* ]] ; } ; then
    write_log debug "Deleting $s3_filename"
    run_as_user aws ... s3 rm ...
else
    write_log debug "Skipping $s3_filename (does not match this backup job's filename prefix)"
fi

This mirrors the matching behaviour of the existing filesystem/blobxfer branches, so retention semantics stay consistent across all three backup locations.

Test plan

• bash -n install/assets/functions/10-db-backup — syntax OK
• Point two backup jobs at the same MinIO/S3 bucket + path prefix and confirm each job's cleanup only removes its own aged backups, leaving the other job's files intact

Validation

Tested locally against a real S3 backend: a MinIO container with the actual (unmodified) cleanup_old_data() function exercised over the live aws s3 ls/rm API. The function was run inside a Linux container (amazon/aws-cli) so the GNU date/aws environment matches production, and the function bodies were extracted from both the pre-fix commit and this fix to confirm the real code paths were exercised. Bucket seeded with two jobs' backups (pgsql_db1_myhost_*, pgsql_db2_myhost_*), a postgres globals file (pgsql_globals_myhost_*), and an unrelated object; retention set so every object is age-eligible, isolating the new prefix filter.

Scenario	Deleted	Survived	Result
Pre-fix, base pgsql_db1_myhost	all 5 (incl. other job + unrelated)	none	bug reproduced
Fixed, base pgsql_db1_myhost, globals off	pgsql_db1_myhost_* (2)	db2, globals, unrelated	✅
Fixed, base pgsql_db1_myhost, globals on	pgsql_db1_myhost_* + globals	db2, unrelated	✅ (matches filesystem-branch semantics)
Fixed, empty filename base	none	all	✅ (empty base does not fall back to match-all)