cmd+rt+selfhost+test: graduate alloc to (*T | nomem) / ([]T | nomem)

Per Hare convention, alloc is a typed builtin that returns a tagged
union carrying nomem as the OOM variant. Callers spell their policy:
`alloc(T)!` aborts on OOM (the old behavior), `alloc(T)?` propagates
when the enclosing fn already returns nomem.

cstage: check builds TY_TAGGED{*T | nomem} (or {[]T | nomem}); cgen
emits AX=tag, DX=ptr per the general tagged-return ABI (the (*T|!void)
nullable-ptr fold gated in ea76ee4 keeps this clean). wwstage cgalloc
mirrors. rt/alloc.s zeroes AX on syscall error so the builtin's null
check sees a clean 0 instead of mmap's -errno leaking through as a
poisoned pointer.

Migration: 3 `!` sites in test/wcc/700_e2e.c, 1 `!` site in
rt/ensure.ww (preserves the pre-existing sizeof bug tracked by #27),
1 `?` site in selfhost/test/tagged_ptr_ret.ww (allocbox exercises
real `?` propagation against a (*T | nomem) return).

130/130 tests green, 994_w6c_ww + 995_self_rebuild stage byte-identity
preserved. Follow-ups #31 (wwstage checkletassign leniency), #32
(wwstage slice-form gap), #33 (tagged_ptr_ret.ww make-test wiring).

rt/alloc.s

@@ -5,6 +5,12 @@
 //
 // We pin to PROT_READ|PROT_WRITE and MAP_PRIVATE|MAP_ANONYMOUS so
 // callers never have to plumb file descriptors through.
+//
+// On mmap failure the raw syscall returns -errno (negative). Task #30
+// graduated the `alloc` builtin to a fallible `(*T | nomem)` /
+// `([]T | nomem)` signature whose cgen branches on a null return, so
+// the failure path here returns 0 instead of a poisoned pointer. The
+// builtin's caller is expected to `!`/`?` the result.
 
 TEXT rt_alloc,$0
 	MOVQ	DI, SI		// arg 1: length = caller's n
@@ -15,6 +21,10 @@ TEXT rt_alloc,$0
 	MOVQ	$0, R9		// arg 5: offset = 0
 	MOVQ	$9, AX		// syscall: mmap
 	SYSCALL
+	CMPQ	$0, AX
+	JGE	rt_alloc_ok
+	XORQ	AX, AX
+rt_alloc_ok:
 	RET
 
 TEXT rt_free,$0