mat32 module adapted from Gonum matrix removed non-easy portable parts (lot of funcs dependent from lapack64, io stuff). modified tolerances for tests since float32 operations have a lower precision. pull requests will be happily considered.
