Ok, here's a new version incorporating feedback so far.
1. Invoke pg_regress directly (no make).
2. Use PG_TEST_EXTRA="wal_consistency_checking" as a way to opt in to
the more expensive test.
3. Use parallel schedule rather than serial. It's faster but also
the non-determinism might discover more things. This required
changing the TAP test max_connections setting from 10 to 25.
4. Remove some extraneous print statements and
check-if-data-is-replicated-using-SELECT tests that are technically
not needed (I had copied those from 001_stream_rep.pl).