Convert d_mat to stride representation

[fredrik-johansson]

The d_mat type still has a rows pointer. It should use stride instead like the other matrix types. Though this will slow down row swaps in LLL, it should speed up arithmetic which probably makes it worth it.

[fredrik-johansson]

Tested this and it gives a small measurable slowdown in LLL. Since d_mat is only used in LLL currently, it might be better to make the rows based representation internal to the LLL module. Not an urgent change, however.