Move SQL tests of pg_stat_io for WAL data to recovery test 029_stats_restart

Three tests in the main regression test suite are proving to not be
portable across multiple runs on a deployed cluster as stats of
pg_stat_io are reset.  Problems happen for tests on:
- Writes of WAL in the init context, when creating a WAL segment.
- Syncs of WAL in the init context, when creating a WAL segment.
- Reads of WAL in the normal context, requiring a WAL record to be read.
For a `make check`, this could rely on the checkpoint record read by the
startup process when starting the cluster, something that is not going
to work for a deployed node.

Two of the three tests are moved to the recovery TAP test
029_stats_restart, where we already check the consistency of stats
data.  The test for syncs is dropped as TAP can run with fsync=off.  The
other two are checked with some data from a freshly-initialized cluster.

Per discussion with Tom Lane, Bertrand Drouvot and Nazir Bilal Yavuz.

Discussion: https://postgr.es/m/915687.1738780322@sss.pgh.pa.us

src/test/regress/sql/stats.sql

@@ -442,20 +442,6 @@ SELECT (current_schemas(true))[1] = ('pg_temp_' || beid::text) AS match
 FROM pg_stat_get_backend_idset() beid
 WHERE pg_stat_get_backend_pid(beid) = pg_backend_pid();
 
--- Test pg_stat_io for WAL in an init context, that should do writes
--- and syncs.
-SELECT sum(writes) AS writes, sum(fsyncs) AS fsyncs
-  FROM pg_stat_io
-  WHERE context = 'init' AND object = 'wal' \gset io_sum_wal_init_
-SELECT :io_sum_wal_init_writes > 0;
-SELECT current_setting('fsync') = 'off'
-  OR :io_sum_wal_init_fsyncs > 0;
-
--- Test pg_stat_io for WAL in a normal context, that should do reads as well.
-SELECT SUM(reads) > 0
-  FROM pg_stat_io
-  WHERE context = 'normal' AND object = 'wal';
-
 -----
 -- Test that resetting stats works for reset timestamp
 -----