Device-specific - Row-wise reduction: Fixed-up the previous commit so that it doesn't break slices

Enqueueing sizes instead of internal_size but performing the check inside the kernel