Fix DimeNet envelope bug with missing clamping

[danielzuegner]

What does this fix?

In the Envelope class used in the radial basis function in DimeNet, there's a bug. More precisely, the forward() is not clamping d/c to 1, as is done in the original implementation. This could lead to exploding output of the Envelope when d/c is around 2 (values go up to 300). This lead to numerical instability for me.

Let me know in case you have any questions.

[codecov]

Codecov Report

Merging #4506 (55b37c3) into master (cae37d3) will not change coverage.
The diff coverage is 0.00%.

❗ Current head 55b37c3 differs from pull request most recent head 3286297. Consider uploading reports for the commit 3286297 to get more accurate results

@@           Coverage Diff           @@
##           master    #4506   +/-   ##
=======================================
  Coverage   82.67%   82.67%           
=======================================
  Files         313      313           
  Lines       16321    16321           
=======================================
  Hits        13493    13493           
  Misses       2828     2828           

Impacted Files	Coverage Δ
torch_geometric/nn/models/dimenet.py	17.66% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update cae37d3...3286297. Read the comment docs.

[rusty1s]

Thanks for fixing :)