Register integral overload of isinf isnan nad isfinite

[JohanMabille]

Currently only the overloads accepting floating point types are registered. This is due to an issue in xsimd that should be fixed in xtensor-stack/xsimd#402