Add FeaturePropagation as a transform

[EdisonLeeeee]

This PR implements the FeaturePropagation method proposed in On the Unreasonable Effectiveness of Feature propagation in Learning on Graphs with Missing Node Features.

Based on the prev discussion in this issue, I implement FeaturePropagation as a transform. A minimal example to reproduce the experiments is also provided.

[wsad1]

Really appreciate you adding this. Please add a test for this feature. Left some initial comments, will take a look again in a few days.

[emalgorithm]

Hi, the author of the paper here 👋 . Thanks a lot for working on this and let me know if you have any question!

[wsad1]

Thanks a lot for adding this. left some final comments.

[EdisonLeeeee]

@wsad1 Thanks for your thorough reviews. Would it be able to merge now?

[rusty1s]

Thank you! Made some minor modifications.

[codecov]

Codecov Report

Merging #5387 (fe33faa) into master (af951b8) will increase coverage by 0.03%.
The diff coverage is 97.29%.

❗ Current head fe33faa differs from pull request most recent head 9690d46. Consider uploading reports for the commit 9690d46 to get more accurate results

@@            Coverage Diff             @@
##           master    #5387      +/-   ##
==========================================
+ Coverage   83.37%   83.40%   +0.03%     
==========================================
  Files         346      347       +1     
  Lines       18805    18842      +37     
==========================================
+ Hits        15679    15716      +37     
  Misses       3126     3126              

Impacted Files	Coverage Δ
torch_geometric/graphgym/config_store.py	81.69% <ø> (+0.65%)	⬆️
torch_geometric/transforms/feature_propagation.py	97.22% <97.22%> (ø)
torch_geometric/transforms/__init__.py	100.00% <100.00%> (ø)

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