Add GNNFF model

[ken2403]

This PR adds an implementation of a graph convolutional neural network model called GNNFF.

• It uses the same inputs as the already implemented DimeNet.
• It is fast because it can find the forces between atoms in a material without automatic differentiation.
• It constructs a force field that is translationally-invariant and rotationally-covariant with respect to the atomic coordinates.

[codecov]

Codecov Report

Merging #5866 (8abb176) into master (9aeebac) will increase coverage by 0.13%.
The diff coverage is 99.13%.

❗ Current head 8abb176 differs from pull request most recent head 85347a4. Consider uploading reports for the commit 85347a4 to get more accurate results

@@            Coverage Diff             @@
##           master    #5866      +/-   ##
==========================================
+ Coverage   84.39%   84.52%   +0.13%     
==========================================
  Files         392      393       +1     
  Lines       21461    21561     +100     
==========================================
+ Hits        18111    18225     +114     
+ Misses       3350     3336      -14     

Impacted Files	Coverage Δ
torch_geometric/nn/models/dimenet.py	17.55% <93.33%> (+3.03%)	⬆️
torch_geometric/nn/models/__init__.py	100.00% <100.00%> (ø)
torch_geometric/nn/models/gnnff.py	100.00% <100.00%> (ø)

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[rusty1s]

Thank you! Sorry for taking me so long to merge this :(

[ken2403]

Thank you for your kind review.
I'm proud to have contributed in any small way to such a great project!

I am one of the users who was fascinated by the tremendous potential of PyG!
I will continue to contribute actively!