pg_dump: Adjust reltuples from 0 to -1 for dumps of older versions.

Before v14, a reltuples value of 0 was ambiguous: it could either
mean the relation is empty, or it could mean that it hadn't yet
been vacuumed or analyzed.  (Commit 3d351d9 taught v14 and newer
to use -1 for the latter case.)  This ambiguity allegedly can cause
the planner to choose inefficient plans after restoring to v18 or
newer.  To fix, let's just dump reltuples as -1 in that case.  This
will cause some truly empty tables to be seen as not-yet-processed,
but that seems unlikely to cause too much trouble in practice.

Note that we could alternatively teach pg_restore_relation_stats()
to translate reltuples based on the version argument, but since
that function doesn't exist until v18, there's no particular
advantage to that approach.  That is, there's no chance of
restoring stats dumped from a pre-v14 server to another pre-v14
server.  Per discussion, the current policy is to fix pre-v18
behavior differences during export and everything else during
import.

Commit 9879105 fixed a similar problem for vacuumdb by removing
the check for reltuples != 0.  Presumably we could reinstate that
check now, but I've chosen to leave it in place in case reltuples
isn't accurate.  As before, processing some empty tables seems
relatively harmless.

Author: Hari Krishna Sunder <hari.db.pg@gmail.com>
Reviewed-by: Jeff Davis <pgsql@j-davis.com>
Reviewed-by: Corey Huinker <corey.huinker@gmail.com>
Discussion: https://postgr.es/m/CAAeiqZ0o2p4SX5_xPcuAbbsmXjg6MJLNuPYSLUjC%3DWh-VeW64A%40mail.gmail.com
(cherry picked from commit 5d6eac8)

Author: nathan-bossart

src/bin/pg_dump/pg_dump.c

@@ -10184,7 +10184,20 @@ dumpRelationStats_dumper(Archive *fout, const void *userArg, const TocEntry *te)
 	appendStringLiteralAH(out, rsinfo->dobj.name, fout);
 	appendPQExpBufferStr(out, ",\n");
 	appendPQExpBuffer(out, "\t'relpages', '%d'::integer,\n", rsinfo->relpages);
-	appendPQExpBuffer(out, "\t'reltuples', '%s'::real,\n", rsinfo->reltuples);
+
+	/*
+	 * Before v14, a reltuples value of 0 was ambiguous: it could either mean
+	 * the relation is empty, or it could mean that it hadn't yet been
+	 * vacuumed or analyzed.  (Newer versions use -1 for the latter case.)
+	 * This ambiguity allegedly can cause the planner to choose inefficient
+	 * plans after restoring to v18 or newer.  To deal with this, let's just
+	 * set reltuples to -1 in that case.
+	 */
+	if (fout->remoteVersion < 140000 && strcmp("0", rsinfo->reltuples) == 0)
+		appendPQExpBufferStr(out, "\t'reltuples', '-1'::real,\n");
+	else
+		appendPQExpBuffer(out, "\t'reltuples', '%s'::real,\n", rsinfo->reltuples);
+
 	appendPQExpBuffer(out, "\t'relallvisible', '%d'::integer\n);\n",
 					  rsinfo->relallvisible);