Fix out-of-bound reads in the LHS packing function

- Rewrite the optimized path for kr = 16
- The out-of-bound check is not required anymore as the optimized path
is only for the in-bound portion of the matrix

Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>

Approved-by: Felix Johnny Thomasmathibalan <felixjohnny.thomasmathibalan@arm.com>