High overloaded mask data type

[SergeyKopienko]

oneDPL/include/oneapi/dpl/pstl/hetero/dpcpp/parallel_backend_sycl.h

Line 770 in 95bedef

oneapi::dpl::__par_backend_hetero::__buffer<int32_t> __mask_buf(__n);

I think it's very memory-expensive to store 4 bytes for each element just as usage flag (a mask).

@mmichel11 @danhoeflinger Could you please take a look?