cinder-manage: Use same timestamp for purging deleted rows

The ``cinder-manage db purge <age_in_days>`` command currently
recalculates the timetamp for deleting rows for each table it
processes. This can lead to foreign key constraint errors, as
secondary (dependent) tables may be deleted before their
corresponding primary (parent) tables, each using slightly
different timestamps.

This patch addresses the issue by calculating the timestamp once
and reusing it across all bulk delete operations, ensuring that
all tables are purged relative to the same point in time.

Closes-Bug: #2111461
Change-Id: I2aa881936b85b3876d6c9c9cfe3b26932f65241a

cinder/db/sqlalchemy/api.py

@@ -8100,6 +8100,7 @@ def purge_deleted_rows(context, age_in_days):
     metadata = MetaData()
     metadata.reflect(engine)
 
+    deleted_age = timeutils.utcnow() - dt.timedelta(days=age_in_days)
     for table in reversed(metadata.sorted_tables):
         if 'deleted' not in table.columns.keys():
             continue
@@ -8110,7 +8111,6 @@ def purge_deleted_rows(context, age_in_days):
             {'age': age_in_days, 'table': table},
         )
 
-        deleted_age = timeutils.utcnow() - dt.timedelta(days=age_in_days)
         try:
             # Delete child records first from quality_of_service_specs
             # table to avoid FK constraints