Implement interface for external allocators

[sbaldu]

This PR adds to the allocation functions the possibility of passing an externally defined allocator like a caching allocator.
Since there is not a universal convention on the interface of device caching allocators, users are expected to wrap their allocators in a class respecting the concepts::CachingAllocator concept.
For example:

template <typename CachingAllocator>
class AllocatorWrapper {
private:
  std::reference_wrapper<CachingAllocator> m_allocator;
  typename CachingAllocator::Queue m_queue;

public:
  AllocatorWrapper(CachingAllocator& allocator, typename CachingAllocator::Queue queue)
      : m_allocator(allocator), m_queue(std::move(queue)) {}

  void* allocate(std::size_t bytes, std::size_t /* align */) {
    return m_allocator.get().allocate(bytes, m_queue);
  }

  void deallocate(void* ptr) { m_allocator.get().free(ptr); }
};

The allocator wrapper can then be constructed and passed to the allocation functions:

AllocatorWrapper<Alloc> wrapper(cachingAlloc, queue);
auto buf = alpaka::allocAsyncBuf<float, Idx>(queue, extent, wrapper);