OneHot encoding of inputs is potentially incorrect, correction required

[RMichae1]

In the pr/solver.py Line 123 all inputs are one-hot encoded.
However this is incorrect for some problem cases for example discrete categorical inputs.
This leads to a wrong optimization scheme ultimately (and a significantly larger problem space).

TODO: only if absolutely required do torch.nn.functional.one_hot(X_init) otherwise keep input variables (integers) as is.