OpenMPTarget Index range fix for MDRange.

[rgayatri23]

This fixes the inconsistency in the OpenMPTarget backend observed here #4912

[dalg24]

I think we should update

kokkos/core/src/OpenMPTarget/Kokkos_OpenMPTarget_Parallel_MDRange.hpp

Lines 140 to 157 in cdcd690

	#else
	const ptrdiff_t begin_0 = offset[0];
	ptrdiff_t end_0 = begin_0 + policy.m_tile[0];
	end_0 = end_0 < policy.m_upper[0] ? end_0 : policy.m_upper[0];
	const ptrdiff_t begin_1 = offset[1];
	ptrdiff_t end_1 = begin_1 + policy.m_tile[1];
	end_1 = end_1 < policy.m_upper[1] ? end_1 : policy.m_upper[1];
	#pragma omp for collapse(2)
	for (ptrdiff_t i0 = begin_0; i0 < end_0; ++i0)
	for (ptrdiff_t i1 = begin_1; i1 < end_1; ++i1) {
	if constexpr (std::is_void<typename Policy::work_tag>::value)
	functor(i0, i1);
	else
	functor(typename Policy::work_tag(), i0, i1);
	}
	#endif

and alike also (code path guarded by #ifndef KOKKOS_IMPL_MDRANGE_USE_NO_TILES )

[dalg24]

Is this

kokkos/core/src/OpenMPTarget/Kokkos_OpenMPTarget_Parallel_MDRange.hpp

Lines 52 to 55 in cdcd690

	// WORKAROUND OPENMPTARGET: sometimes tile sizes don't make it correctly,
	// this was tracked down to a bug in clang with regards of mapping structs
	// with arrays of long in it. Arrays of int might be fine though ...
	#define KOKKOS_IMPL_MDRANGE_USE_NO_TILES // undef EOF

still an issue?

[masterleinad]

This is progress.

[dalg24]

This is progress.

Agreed. OK I'll merge as is and we can fix the alternate code path as a follow-up.