Support for rational and floating-point matrices

[brainandforce]

This package assumes the user is interested in finding the Smith normal form of Integer matrices, but this thread indicates interest in applying the Smith normal form to matrices of Rational or AbstractFloat entries. We should provide support for this.

[brainandforce]

Whenever a matrix is given with elements that subtype Real but not Integer, the default behavior is to convert the matrix to a matrix of Int arguments and then attempt the decomposition.

I'm thinking for floating point matrices, the ideal treatment is to try to convert the arguments to Rational{Int} entries, perhaps with a common denominator. The issue with this is that for an arbitrary matrix, the common denominator may be enormous, and that would likely overflow.

For the sake of type stability, the documentation should encourage users to convert matrices to Integer matrices if possible using (::Type{<:Integer}).(::AbstractMatrix) or round(::Type{<:Integer}, ::AbstractMatrix) if integer results are desired. The Smith and Hermite normal form functions should be type-stable.