Replace kernel computation in kgof with hyppo's compute_kern

[darsh-patel]

Is your feature request related to a problem? Please describe.
Currently, the kernel goodness-of-fit library kgof has its own unique implementation of a kernel evaluation due to a unique data structure returned by autograd.numpy gradient calculations. This ArrayBox data structure does not share properties with numpy arrays and therefore must be handled separately and cannot be fed into compute_kern.

Describe the solution you'd like
Ideally, a new implementation would rip and replace kernel.py by removing its kernel evaluation functions and reducing the corresponding dependencies on autograd.numpy to do so. Should be a quick fix if handling of an ArrayBox can be improved or a better gradient calculation implementation is discovered.

Describe alternatives you've considered
There has already been an attempt to integrate compute_kern by replacing the eval functions in kernel.py with instances of compute_kern, but this raises an error in which values in an ArrayBox cannot be reached.

Additional context (e.g. screenshots)