The sta_repo_push_ref table is unnecessarily populated on mirrors

Issue Summary

The sta_repo_push_ref table serves no purpose on mirror nodes and should not be populated. There are a number of negative consequences of populating this table on mirrors:

The table can grow to be quite large, on large instances many tens of millions of rows. This has a large on-disk footprint, potentially many gigabytes in size. The larger downside however is that the use of the H2 database on mirrors is based on the idea that mirror nodes will generally not grow large tables.

A specific example of the downsides of this is the Liquibase upgrade task with id BBSDEV-21559-2 (applied as part of upgrading to of through 7.1). Recent experience on an internal Atlassian mirror where (for 17M rows) an upgrade from 6.10 to 7.17 ran for many hours before eventually being abandoned and reinitialising and resyncing the mirror.

Expected Results

The sta_repo_push_ref table is of zero length on mirror nodes.

Workaround

Truncate the sta_repo_push_ref which may take an extremely long time, or rebuild the mirror.