Ignore other pg_repack clients in SQL_XID_SNAPSHOT

[schmiddy]

Thibaud WALKOWIAK complained recently about pg_repack clients having to wait on each other while concurrently operating on tables in the same database. Presumably this wait comes from the SQL_XID_SNAPSHOT check, which does not bother ignoring other pg_repack clients based on application_name.

I proposed a simple fix to have SQL_XID_SNAPSHOT ignore other pg_repack clients using pg_stat_activity.application_name here:
http://lists.pgfoundry.org/pipermail/reorg-general/2012-November/000137.html

(Will only help on 9.0 or later, since that's when we have application_name exposed.)

[schmiddy]

Fixed in a072cc9 and f6ca290. The queries ended up tiny bit different than what I initially proposed:

• make sure to explicitly exclude conn2 based on its backend PID, since we can't trust application_name in all cases (e.g. under installcheck, or on old versions)
• use LEFT JOIN against pg_stat_activity, in case there are locks in pg_locks which don't show up in pg_stat_activity
• don't hardcode the application_name 'pg_repack' in the query, just in case our PROGRAM_NAME is something different