Use FlatPageMap on OS with lazy commit

If the operating system will allocate private pages on demand for the
pagemap then use the FlatPageMap by default as it generates better code
for deallocation.

src/mem/alloc.h

@@ -46,8 +46,10 @@ namespace snmalloc
 // Use flat map is under a single node.
 #  define SNMALLOC_MAX_FLATPAGEMAP_SIZE PAGEMAP_NODE_SIZE
 #endif
-  static constexpr bool USE_FLATPAGEMAP = SNMALLOC_MAX_FLATPAGEMAP_SIZE >=
-    sizeof(FlatPagemap<SUPERSLAB_BITS, uint8_t>);
+  static constexpr bool USE_FLATPAGEMAP =
+    (Pal::pal_features & PalFeatures::LazyCommit) ||
+    (SNMALLOC_MAX_FLATPAGEMAP_SIZE >=
+     sizeof(FlatPagemap<SUPERSLAB_BITS, uint8_t>));
 
   using SuperslabPagemap = std::conditional_t<
     USE_FLATPAGEMAP,
@@ -1152,6 +1154,8 @@ namespace snmalloc
     SNMALLOC_SLOW_PATH void small_dealloc_offseted_slow(
       Superslab* super, void* p, sizeclass_t sizeclass)
     {
+      handle_message_queue();
+
       bool was_full = super->is_full();
       SlabList* sl = &small_classes[sizeclass];
       Slab* slab = Slab::get(p);