Introdue a type `alpaka::warp::mask_t`

[fwyzard]

The type of the value returned by warp operations like ballot() and activemask() is different for different back-ends.

It would help to write generic code if alpaka would provide an alias for this type, for example

namespace alpaka::warp {
  template <typename TWarp>
  using mask_t = ...;
}

[fwyzard]

Note that HIP uses uint64_t (actually unsigned long long) independently of the GPU wavefront size:

unsigned long long __ballot(int predicate);
unsigned long long __activemask();

[fwyzard]

Note that oneAPI uses a struct sub_group_mask with an implementation-defined size, independent from the backend.
The current code internally uses an uint64_t.