Avoid repeated creation/freeing of per-subre DFAs during regex search.

In nested sub-regex trees, lower-level nodes created DFAs and then
destroyed them again before exiting, which is a bit dumb considering that
the recursive search is likely to call those nodes again later.  Instead
cache each created DFA until the end of pg_regexec().  This is basically a
space for time tradeoff, in that it might increase the maximum memory
usage.  However, in most regex patterns there are not all that many subre
nodes, so not that many DFAs --- and in any case, the peak usage occurs
when reaching the bottom recursion level, and except for alternation cases
that's going to be the same anyway.

src/include/regex/regguts.h

@@ -409,7 +409,7 @@ struct subre
 #define  PREF(f) ((f)&LOCAL)
 #define  PREF2(f1, f2)	 ((PREF(f1) != 0) ? PREF(f1) : PREF(f2))
 #define  COMBINE(f1, f2) (UP((f1)|(f2)) | PREF2(f1, f2))
-	short		id;				/* ID of subre (1..ntree) */
+	short		id;				/* ID of subre (1..ntree-1) */
 	int			subno;			/* subexpression number (for 'b' and '(') */
 	short		min;			/* min repetitions for iteration or backref */
 	short		max;			/* max repetitions for iteration or backref */
@@ -446,7 +446,7 @@ struct guts
 	size_t		nsub;			/* copy of re_nsub */
 	struct subre *tree;
 	struct cnfa search;			/* for fast preliminary search */
-	int			ntree;
+	int			ntree;			/* number of subre's, less one */
 	struct colormap cmap;
 	int			FUNCPTR(compare, (const chr *, const chr *, size_t));
 	struct subre *lacons;		/* lookahead-constraint vector */