wcc: defer capacity 32 with a loud cap error, both stages

wwstage capped defers at 16 and SILENTLY DROPPED the 17th; cstage
capped at 32. Align the cap at 32 and make exceeding it a loud
compile error in BOTH stages — the silent 16-vs-32 split was the bug
(a defer that never runs is a leaked resource). Both stages move in
one commit: one cap contract.

selfhost/cmd/wwdump/main.combined.ww

@@ -35682,10 +35682,17 @@ fn cgstmt(c: *cgen, n: *node) void = {
 	if (k == nkind.N_YIELD) { cgyield(c, n); return; };
 
 	if (k == nkind.N_DEFER) {
-		if (c.defertop < DEFER_MAX) {
-			c.deferbuf[c.defertop] = n.lhs;
-			c.defertop += 1;
+		// #40: at the cap, fail loud in BOTH stages rather than
+		// silently drop the deferred call. cstage's DEFER_MAX was 32
+		// and also dropped silently past it; the runtime-correct target
+		// is a hard stop at the shared cap (cgen.c twin fatals too).
+		if (c.defertop >= DEFER_MAX) {
+			let msg: str = "cgen: too many defers in one function\n";
+			os.write(2, msg.ptr, msg.len: u64);
+			os.exit(1);
 		};
+		c.deferbuf[c.defertop] = n.lhs;
+		c.defertop += 1;
 		return;
 	};
 
@@ -41060,7 +41067,7 @@ type enumtype = struct {
 };
 
 def LOOP_MAX: i32 = 16;
-def DEFER_MAX: i32 = 16;
+def DEFER_MAX: i32 = 32;	// #40: match cstage cgen.c DEFER_MAX (shared cap)
 
 // The SysV register-return-ABI caps — the SINGLE SSoT shared by the sret
 // classifier (sretretsize over-cap-tuple arm) AND every emit/receive site