Remove unnecessary inclusion of self-loops when sampling negative edges

[gengala]

Handling self-loops is only useful when negative edges are not provided.

[codecov]

Codecov Report

Merging #4880 (c2c77f1) into master (b32bbdb) will increase coverage by 0.05%.
The diff coverage is n/a.

❗ Current head c2c77f1 differs from pull request most recent head 3973fef. Consider uploading reports for the commit 3973fef to get more accurate results

@@            Coverage Diff             @@
##           master    #4880      +/-   ##
==========================================
+ Coverage   82.70%   82.75%   +0.05%     
==========================================
  Files         329      329              
  Lines       17846    17707     -139     
==========================================
- Hits        14760    14654     -106     
+ Misses       3086     3053      -33     

Impacted Files	Coverage Δ
torch_geometric/nn/models/autoencoder.py	100.00% <ø> (ø)
torch_geometric/data/in_memory_dataset.py	77.27% <0.00%> (-3.89%)	⬇️
torch_geometric/data/separate.py	97.77% <0.00%> (-2.23%)	⬇️
torch_geometric/loader/link_neighbor_loader.py	91.85% <0.00%> (-0.69%)	⬇️
torch_geometric/data/data.py	91.43% <0.00%> (-0.14%)	⬇️
torch_geometric/data/hetero_data.py	94.13% <0.00%> (-0.09%)	⬇️
torch_geometric/data/__init__.py	100.00% <0.00%> (ø)
torch_geometric/nn/dense/mincut_pool.py	100.00% <0.00%> (ø)
torch_geometric/loader/neighbor_loader.py	86.44% <0.00%> (+0.27%)	⬆️
torch_geometric/data/collate.py	96.69% <0.00%> (+0.29%)	⬆️
... and 3 more

---

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 b32bbdb...3973fef. Read the comment docs.

[rusty1s]

Thanks for the PR. Meanwhile, negative_sampling will not include self-loops in the computation, so we can drop the removal and addition of self-loops here completely. WDYT?

[gengala]

You are right :)