Fix parallel index scan hang with deleted or half-dead pages.

The previous coding forgot to release the scan before seizing
it again, leading to a lockup.

Report by Patrick Hemmer.  Diagnosis by Thomas Munro.  Patch by
Amit Kapila.

Discussion: http://postgr.es/m/CAEepm=2xZUcOGP9V0O_G0=2P2wwXwPrkF=upWTCJSisUxMnuSg@mail.gmail.com

Author: robertmhaas

src/backend/access/nbtree/nbtsearch.c

@@ -1486,6 +1486,11 @@ _bt_readnextpage(IndexScanDesc scan, BlockNumber blkno, ScanDirection dir)
 				if (_bt_readpage(scan, dir, P_FIRSTDATAKEY(opaque)))
 					break;
 			}
+			else if (scan->parallel_scan != NULL)
+			{
+				/* allow next page be processed by parallel worker */
+				_bt_parallel_release(scan, opaque->btpo_next);
+			}
 
 			/* nope, keep going */
 			if (scan->parallel_scan != NULL)
@@ -1581,6 +1586,11 @@ _bt_readnextpage(IndexScanDesc scan, BlockNumber blkno, ScanDirection dir)
 				if (_bt_readpage(scan, dir, PageGetMaxOffsetNumber(page)))
 					break;
 			}
+			else if (scan->parallel_scan != NULL)
+			{
+				/* allow next page be processed by parallel worker */
+				_bt_parallel_release(scan, BufferGetBlockNumber(so->currPos.buf));
+			}
 
 			/*
 			 * For parallel scans, get the last page scanned as it is quite