Change convention for the stride of the rhs_packed matrix

Before the rhs_packed_stride depended on the nr value. With this patch,
we remove this dependency so this stride can be used to calculate the
offset for the N values stored in the packed matrix without dividing N
by nr

Resolves COMPMID-7319

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