Ensure all nmod-type functions works with intended range of modulo

[albinahlback]

For instance, nmod_mat_one does not work with modulo 1. I'm sure that this is the only modulus that is problematic.

Personally, I don't think we should allow modulo 1 since that only yield trivial results. Not sure how wide-spread this is.

Edit: nmod_mat_randfull is also problematic.

[albinahlback]

We should implement functions such as nmod_mat_is_canonical, just like fmpz_is_canonical.

[fredrik-johansson]

Personally, I don't think we should allow modulo 1 since that only yield trivial results.

It is useful, just like it is useful to allow the 0 x 0 matrix, empty set, etc. For example, a user who CRTs results mod m1, m2, m3, ... might start with 1 as an initial value.