Respect flow argument in GCN normalization

[gau-nernst]

Add flow to gcn_norm to correctly normalize adjacency matrix when flow=target_to_source for directed graphs. This is discussed in #5100.

A simple test is added to verify this new behavior.

@rusty1s Let me know if you need me to edit any thing.

[codecov]

Codecov Report

Merging #5149 (4d9f4cb) into master (8f27635) will increase coverage by 0.00%.
The diff coverage is 100.00%.

❗ Current head 4d9f4cb differs from pull request most recent head c5be455. Consider uploading reports for the commit c5be455 to get more accurate results

@@           Coverage Diff           @@
##           master    #5149   +/-   ##
=======================================
  Coverage   82.98%   82.98%           
=======================================
  Files         333      333           
  Lines       18365    18367    +2     
=======================================
+ Hits        15240    15242    +2     
  Misses       3125     3125           

Impacted Files	Coverage Δ
torch_geometric/nn/conv/gcn_conv.py	97.89% <100.00%> (+0.04%)	⬆️

📣 Codecov can now indicate which changes are the most critical in Pull Requests. Learn more

[wsad1]

Thanks for the fix.

[gau-nernst]

@wsad1 I have updated the changelog and merged from master. 1 unrelated test is failing (test_copy_linear[True]), probably due to nan values.

[rusty1s]

Thanks @gau-nernst. Also added the flow argument to other layers that utilize gcn_norm :)