[SYCL] Fix for subbuffer

[dyniols]

This PR fixes test-e2e/Basic/buffer/subbuffer.cpp test. It was failing Gen12 OpenCL. #15151

Looks like we incorrectly create subbuffer with size which is smaller than anticipated.

Example

  std::size_t size = 10;
  std::size_t offset = size / 2, subbuf_size = 5, offset_inside_subbuf = 2,
              subbuffer_access_range = subbuf_size - offset_inside_subbuf;
  std::vector<int> vec(size);
  std::iota(vec.begin(), vec.end(), 0);

{
    sycl::buffer<int, 1> buf(vec.data(), size);
    sycl::buffer<int, 1> subbuf(buf, sycl::id<1>(offset),
                                sycl::range<1>(subbuf_size));  
    q.submit([&](sycl::handler &cgh) {
      auto acc = subbuf.get_access<sycl::access::mode::read_write>(
          cgh, sycl::range<1>(subbuffer_access_range),
          sycl::id<1>(offset_inside_subbuf));
      cgh.parallel_for(sycl::range<1>(subbuffer_access_range),
                                     [=](sycl::id<1> i) { 
                                      acc[i] *= -1; });
    });
}

Expected result is 0 1 2 3 4 5 6 -7 -8 -9 but without this change result is 0 1 2 3 4 5 6 -7 8 9

However when MRequirement.MAccessRange is replaced with MRequirement.MMemoryRange we create subbuffer of size just 5 instead of 3 and expected result is valid.

Here is tracker explaining why it happens on Gen12: CMPLRLLVM-72513

[cperkinsintel]

finally

[github-actions]

@intel/llvm-gatekeepers please consider merging

[dyniols]

@intel/llvm-gatekeepers Can we merge it ? Thanks :)