Allow custom allocator for DistributedDomain and LocalDomain

[cwpearson]

To allow for easier testing and extension to future CPU-only stencils, we should define a DeviceAllocator and a ManagedAllocator along the lines of http://www.josuttis.com/cppcode/myalloc.hpp, then, modify XxxDomain to use the allocator.

The new declaration would be like

template<typename Alloc = DeviceAllocator<char>>
class LocalDomain {
...
};

A challenge is that the classes derived from HaloSenders currently have a LocalDomain * member, which means they would have to be templated around both Sender and LocalDomain. They can still derive from HaloSender, though, so the DistributedDomain should be okay.

[cwpearson]

Allocators were added in 0749943