Hessian evaluation fails for Python-native Potential classes

[adrn]

As reported by @eugvas via email

For the two Python-native classes (Kuzmin and HarmonicOscillator), the hessian seems to fail outright with some complaint about the array shape: cannot reshape array of size 1 into shape (3,3,1) and the Kuzmin potential even fails to compute the potential or gradient for a single point input, also complaining about the shape: not enough values to unpack (expected 3, got 1)). So I could not test the compatibility with pure python potentials in gala.