Refactor the packing function for non-transposed RHS

- Refactor the non-transposed case. Now, if the RHS matrix is not
transposed, each byte holds two N values
- Extend the example to support the NxK and KxN RHS matrices
- Add support for non-even K values

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