Update code for new numpy/pyg/cuda dependencies

[jgbrasier]

1. Updated .gitignore to be more inclusive
2. Updated working core dependencies to new versions
3. Added *args, **kwargs to __inc__ and __cat_dim__
4. Updated shape operator estimation to support new numpy
5. Added pykeops test file

[rubenalv]

in geometry_processing.py
torch.solve(B,A).solution
becomes (with torch > 1.9)
torch.linalg.solve(A,B) # note the argument swap, see details in pyg-team/pytorch_geometric#7670

But your code reads as
torch.linalg.solve(B,A) # with no argument swap, so the result will be different

Eg.
`

A = torch.tensor([[6.80, -2.11, 5.66, 5.97, 8.23],
... [-6.05, -3.30, 5.36, -4.44, 1.08],
... [-0.45, 2.58, -2.70, 0.27, 9.04],
... [8.32, 2.71, 4.35, -7.17, 2.14],
... [-9.67, -5.14, -7.26, 6.08, -6.87]]).t()
B = torch.tensor([[6.80, -2.11, 5.66, 5.97, 8.23],
... [-6.05, -3.30, 5.36, -4.44, 1.08],
... [-0.45, 2.58, -2.70, 0.27, 9.04],
... [8.32, 2.71, 4.35, -7.17, 2.14],
... [-9.67, -5.14, -7.26, 6.08, 0]]).t()
X = torch.linalg.solve(A, B)
X1 = torch.linalg.solve(B,A)
X
tensor([[ 1.0000e+00, 2.8969e-08, -2.7807e-09, 3.0076e-09, 2.2726e-01],
[ 2.1641e-09, 1.0000e+00, 1.3598e-07, -4.7159e-08, 2.0627e-01],
[ 1.6681e-08, 4.7242e-08, 1.0000e+00, -8.4797e-08, 6.6163e-01],
[-1.9888e-08, 1.3481e-07, -6.8328e-08, 1.0000e+00, 3.1942e-01],
[-0.0000e+00, 1.3905e-07, -6.9526e-08, -2.0858e-07, 1.2748e+00]])
X1
tensor([[ 1.0000e+00, 7.8218e-09, 0.0000e+00, 2.0401e-08, -1.7827e-01],
[ 2.1641e-09, 1.0000e+00, -0.0000e+00, -7.1953e-08, -1.6181e-01],
[ 1.6681e-08, -2.4927e-08, 1.0000e+00, 4.6463e-08, -5.1901e-01],
[-1.9888e-08, 9.9970e-08, -5.0908e-08, 1.0000e+00, -2.5056e-01],
[-0.0000e+00, 1.0908e-07, -5.4539e-08, -1.6362e-07, 7.8444e-01]])
`