Add GATv3Conv implementation and tests

[davidnarganes]

Description
This Pull Request introduces GATv3Conv, a context-aware extension of GATv2. It applies element-wise multiplication of source and target feature embeddings, incorporates a learnable context attention layer, and supports optional weight sharing. Inspired by the GATher framework (Narganes-Carlon et al., 2024).

Key Changes

• New file: torch_geometric/nn/conv/gatv3_conv.py – Implements the GATv3Conv layer.
• Tests: test/nn/conv/test_gatv3_conv.py – Validates functionality, including bipartite inputs, edge attributes, and return of attention weights.
• Initialization: torch_geometric/nn/conv/__init__.py – Exports GATv3Conv for usage throughout the PyG library.

Reference

• GATher: Graph Attention Based Predictions of Gene-Disease Links
  https://arxiv.org/abs/2409.16327

This work is authored by David Narganes-Carlon and collaborators. Let me know if further modifications or clarifications are needed.

[puririshi98]

this is good at a high level. however i want to see how it compares to existing work. Can you please update this example:
https://github.com/pyg-team/pytorch_geometric/blob/master/examples/ogbn_train.py#L31
to have a "--gnn-choice" arg parse option, with choices ["sage, gat, gatv3"]. and run all 3 in your environment to see how they compare. Please make the highest test acc the default. I can review a little closer once that initial test is done

[puririshi98]

also please make sure the CI (automated testing) is all green