Fix pg_visibility regression failure with CLOBBER_CACHE_ALWAYS

Commit 8e03eb92 reverted a bit too much code, reintroducing one of the
issues fixed by 39b66a91 - a page might have been left partially empty
after relcache invalidation.

Reported-By: Tom Lane
Author: Masahiko Sawada
Discussion: https://postgr.es/m/822752.1623032114@sss.pgh.pa.us
Discussion: https://postgr.es/m/CAD21AoA%3D%3Df2VSw3c-Cp_y%3DWLKHMKc1D6s7g3YWsCOvgaYPpJcg%40mail.gmail.com

src/backend/access/heap/hio.c

@@ -407,19 +407,19 @@ RelationGetBufferForTuple(Relation relation, Size len,
 		 * target.
 		 */
 		targetBlock = GetPageWithFreeSpace(relation, targetFreeSpace);
+	}
 
-		/*
-		 * If the FSM knows nothing of the rel, try the last page before we
-		 * give up and extend.  This avoids one-tuple-per-page syndrome during
-		 * bootstrapping or in a recently-started system.
-		 */
-		if (targetBlock == InvalidBlockNumber)
-		{
-			BlockNumber nblocks = RelationGetNumberOfBlocks(relation);
+	/*
+	 * If the FSM knows nothing of the rel, try the last page before we
+	 * give up and extend.  This avoids one-tuple-per-page syndrome during
+	 * bootstrapping or in a recently-started system.
+	 */
+	if (targetBlock == InvalidBlockNumber)
+	{
+		BlockNumber nblocks = RelationGetNumberOfBlocks(relation);
 
-			if (nblocks > 0)
-				targetBlock = nblocks - 1;
-		}
+		if (nblocks > 0)
+			targetBlock = nblocks - 1;
 	}
 
 loop: