SparseTensor support for trim_to_layer

[andreazanetti]

It provides support for SparseTensor data format for the trim_to_layer.py functionality.
The trimming function is isolated in a separate function, but still contained in trim_to_layer.py, as it is useful to that only.
Tests are included extending the test file used for trim_to_layer.py.
Some tests already present in a previous PR were slightly modified as to simulate a adj_t matrix that originates from a BFS traversal of a graph, starting the BFS traversal from the target node (assuming Batch Size = 1)

[codecov]

Codecov Report

Merging #7089 (a06feb8) into master (70cff06) will decrease coverage by 0.47%.
The diff coverage is 100.00%.

❗ Current head a06feb8 differs from pull request most recent head 74b4810. Consider uploading reports for the commit 74b4810 to get more accurate results

@@            Coverage Diff             @@
##           master    #7089      +/-   ##
==========================================
- Coverage   91.72%   91.26%   -0.47%     
==========================================
  Files         437      436       -1     
  Lines       23983    23949      -34     
==========================================
- Hits        21999    21857     -142     
- Misses       1984     2092     +108     

Impacted Files	Coverage Δ
torch_geometric/utils/__init__.py	100.00% <100.00%> (ø)
torch_geometric/utils/trim_to_layer.py	100.00% <100.00%> (ø)

... and 30 files with indirect coverage changes

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

[andreazanetti]

@rusty1s
with reference to the failing tests, I have some questions as it is not clear to me how we would like to manage the absence of torch_sparse in the requirements of PyG.

To my understanding, that absence generates issues with this contribution in a number of tests:
-docs/readthedocs.org:pytorch-geometric — Read the Docs build failed! it fails as torch_sparse is missing

• Testing PyTorch nightly / latest_pytest (pull_request) [Failing after 2m ](https://github.com/pyg-team/pytorch_geometric/actions/runs/4574073196/jobs/8075260403?pr=7089 it fails for the same reason)
• the same holds true for other failures above like Minimal Installation / import (pull_request) and make_html

the only one not impacted by the absence of torch_sparse is the changelog failure, which was however present also in other PRs recently merged... so I guess that should not be blocking for now.

How would you like me to go on?
Should I try to make the whole contribution conditional to the presence (=already installed) of torch_sparse in the user's env?
Thank you!!

[rusty1s]

You need to import from torch_geometric.typing for SparseTensor dependency, and any tests need to be gated behind @withPackage('torch_sparse'). Hope this clarifies your doubts.

[andreazanetti]

You need to import from torch_geometric.typing for SparseTensor dependency, and any tests need to be gated behind @withPackage('torch_sparse'). Hope this clarifies your doubts.

Sure, it clarifies, thanks!