[core] Naive atomic type for ARMv5

std::atomic<bool> is implemented lock free which doesn't work
on ARMv5 and seems like the compiler is generating bogus code.

This workaround is a naive implementation using regular mutexes.

platform/android/src/async_task.cpp

@@ -1,9 +1,9 @@
 #include "run_loop_impl.hpp"
 
 #include <mbgl/util/async_task.hpp>
+#include <mbgl/util/atomic.hpp>
 #include <mbgl/util/run_loop.hpp>
 
-#include <atomic>
 #include <functional>
 
 namespace mbgl {
@@ -45,7 +45,7 @@ private:
 
     // TODO: Use std::atomic_flag if we ever drop
     // support for ARMv5
-    std::atomic<bool> queued;
+    util::Atomic<bool> queued;
     std::function<void()> task;
 };