WALRedo: Limit logging to log_level = ERROR and above (#5587)

This fixes issues in pageserver's walredo process where WALRedo
logs of loglevel=LOG are interpreted as errors.

## Problem

See #5560

## Summary of changes

Set the log level to something that doesn't include LOG.

pgxn/neon_walredo/walredoproc.c

@@ -201,6 +201,16 @@ WalRedoMain(int argc, char *argv[])
 #endif
 
 	am_wal_redo_postgres = true;
+	/*
+	 * Pageserver treats any output to stderr as an ERROR, so we must
+	 * set the log level as early as possible to only log FATAL and 
+	 * above during WAL redo (note that loglevel ERROR also logs LOG,
+	 * which is super strange but that's not something we can solve
+	 * for here. ¯\_(-_-)_/¯
+	 */
+	SetConfigOption("log_min_messages", "FATAL", PGC_SUSET, PGC_S_OVERRIDE);
+	SetConfigOption("client_min_messages", "ERROR", PGC_SUSET,
+					PGC_S_OVERRIDE);
 
 	/*
 	 * WAL redo does not need a large number of buffers. And speed of

test_runner/fixtures/neon_fixtures.py

@@ -1631,7 +1631,6 @@ class NeonPageserver(PgProtocol):
             ".*took more than expected to complete.*",
             # these can happen during shutdown, but it should not be a reason to fail a test
             ".*completed, took longer than expected.*",
-            '.*registered custom resource manager \\\\"neon\\\\".*',
             # AWS S3 may emit 500 errors for keys in a DeleteObjects response: we retry these
             # and it is not a failure of our code when it happens.
             ".*DeleteObjects.*We encountered an internal error. Please try again.*",