Reduce dependence on C++ runtime

If the external thread statics are used, then
we don't need to include some C++ runtime
concepts. This refactoring moves some global initialization under
conditional compilation.

src/test/func/fixed_region/fixed_region.cc

@@ -19,6 +19,8 @@ extern "C" const void* __oe_get_heap_end()
 
 extern "C" void* oe_memset(void* p, int c, size_t size)
 {
+  std::cout << "Memset " << p << " - " << size << std::endl;
+
   return memset(p, c, size);
 }
 
@@ -30,7 +32,7 @@ extern "C" void oe_abort()
 using namespace snmalloc;
 int main()
 {
-  MemoryProviderStateMixin<DefaultPal> mp;
+  auto& mp = *MemoryProviderStateMixin<DefaultPal>::make();
 
   // 28 is large enough to produce a nested allocator.
   // It is also large enough for the example to run in.

src/test/func/thread_alloc_external/thread_alloc_external.cc

@@ -0,0 +1,40 @@
+
+#define SNMALLOC_EXTERNAL_THREAD_ALLOC
+#include <mem/globalalloc.h>
+using namespace snmalloc;
+
+class ThreadAllocUntyped
+{
+public:
+  static void* get()
+  {
+    static thread_local void* alloc = nullptr;
+    if (alloc != nullptr)
+    {
+      return alloc;
+    }
+ 
+    alloc = current_alloc_pool()->acquire();
+    return alloc;
+  }
+};
+
+#include <override/malloc.cc>
+
+int main()
+{
+  MemoryProviderStateMixin<DefaultPal> mp;
+
+  // 28 is large enough to produce a nested allocator.
+  // It is also large enough for the example to run in.
+  // For 1MiB superslabs, SUPERSLAB_BITS + 4 is not big enough for the example.
+ 
+  auto a = ThreadAlloc::get();
+
+  for (size_t i = 0; i < 1000; i++)
+  {
+    auto r1 = a->alloc(100);
+ 
+    free(r1);
+  }
+}

src/test/perf/low_memory/low-memory.cc

@@ -57,7 +57,7 @@ bool has_pressure()
     return false;
   }
 
-  uint64_t current_epoch = default_memory_provider.low_memory_epoch();
+  uint64_t current_epoch = default_memory_provider().low_memory_epoch();
   bool result = epoch != current_epoch;
   epoch = current_epoch;
   return result;