PyTorch Sparse Tensor support: LEConv, LGConv, NNConv, PANConv, SignedConv, and WLConv

[EdisonLeeeee]

No description provided.

[codecov]

Codecov Report

Merging #6936 (1da0fc5) into master (d3043d8) will increase coverage by 0.01%.
The diff coverage is 94.44%.

❗ Current head 1da0fc5 differs from pull request most recent head 07fd9c1. Consider uploading reports for the commit 07fd9c1 to get more accurate results

@@            Coverage Diff             @@
##           master    #6936      +/-   ##
==========================================
+ Coverage   91.43%   91.45%   +0.01%     
==========================================
  Files         431      431              
  Lines       23446    23453       +7     
==========================================
+ Hits        21439    21449      +10     
+ Misses       2007     2004       -3     

Impacted Files	Coverage Δ
torch_geometric/utils/sparse.py	91.78% <ø> (+5.47%)	⬆️
torch_geometric/nn/conv/pan_conv.py	96.77% <85.71%> (-1.51%)	⬇️
torch_geometric/nn/conv/signed_conv.py	100.00% <100.00%> (ø)
torch_geometric/nn/conv/wl_conv.py	97.67% <100.00%> (+0.11%)	⬆️

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

[wsad1]

It looks like we are making the assumption that if is_torch_sparse_tensor then the tensor is of format coo. Like for example in wl_conv we use from_torch_sparse_coo_tensor.