cmd+selfhost+test: gate alloc builtin behind same-module fn alloc

Mirrors the existing abort/assert gates in cstage check.c (strict
same-module lookup rather than scope_lookup_prefer, since lib/os.alloc
under a `use os;` import must not suppress the bare-alloc builtin in
client code). cgen.c shadows the resolution: only fire the rt_alloc
path when the typer left N_CALL.lhs->type == ty_err. wwstage gets a
new samemodfn helper for the matching gate.

Test fixtures: package-main repair for the 3 alloc rows in 700_e2e.c
that the parser was inheriting curmod="os" from the concat'd os.ww;
new shadow-test row asserts a same-module `fn alloc(n: i64) i64`
beats the builtin in cgen.

selfhost/cmd/wcc/cgenexpr.ww

@@ -2938,10 +2938,19 @@ fn cgcall(c: *cgen, n: *node) void = {
 			// value's bytes. For struct literals, lower to rt_alloc
 			// + per-field stores. Mirrors cmd/w6c/cgen.c's N_CALL
 			// alloc path.
+			//
+			// Same-module-scope guard: skip the builtin when a fn
+			// `alloc` is declared in the current module (lib/os and
+			// rt/ensure both shadow it). Mirrors cstage check.c's
+			// scope_lookup_prefer gating on the `abort` precedent;
+			// without it, the bare same-module call lands in the
+			// typed-builtin path and shadows the user decl. Task #23.
 			if (streq(callee.str, "alloc")) {
 				if (n.list != nil) {
-					cgalloc(c, n);
-					return;
+					if (!samemodfn(c, "alloc")) {
+						cgalloc(c, n);
+						return;
+					};
 				};
 			};
 		};