Added batching in transductive setting

[Coerulatus]

Hello everyone,

I have added the possibility of batching the data in the transductive setting.
When working with large graphs, selecting a subset of the graph while keeping the model's performance unchanged for the desired nodes can drastically reduce the memory requirements during training and inference.
In torch_geometric, the NeighborLoader performs neighbor sampling to achieve this. This can be done because, in the normal message-passing framework, the information propagates only as far as the number of message-passing steps performed.
The newly added NeighborCellsLoader works similarly but it also selects the relevant higher-order cells, by sequentially reducing all the incidences.
In the loader, you can also specify the rank to consider, meaning that you can perform batching over the nodes, edges, or any higher-order cell.

I have also added a tutorial that shows the basic functionality of NeighborCellsLoader. It also tests that the approach works as expected by comparing the model's outputs working with the full graph or with the batched one. Interestingly the number of hops needed is not necessarily equal to the number of layers in the higher-order networks. Information, at each layer, can in general travel further than the 1-neighborhood when working with these models.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[codecov]

Codecov Report

Attention: Patch coverage is 92.72727% with 16 lines in your changes missing coverage. Please review.

Project coverage is 90.16%. Comparing base (9271ec4) to head (3f44880).
Report is 9 commits behind head on main.

Files with missing lines	Patch %	Lines
topobenchmark/data/batching/cell_loader.py	85.18%	8 Missing ⚠️
topobenchmark/data/batching/utils.py	94.95%	6 Missing ⚠️
...pobenchmark/data/batching/neighbor_cells_loader.py	96.15%	1 Missing ⚠️
topobenchmark/nn/readouts/propagate_signal_down.py	50.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #128      +/-   ##
==========================================
+ Coverage   89.51%   90.16%   +0.65%     
==========================================
  Files         126      130       +4     
  Lines        3518     3732     +214     
==========================================
+ Hits         3149     3365     +216     
+ Misses        369      367       -2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.