Refactor pyg modules to correctly utilise EdgeIndex

[wsad1]

🛠 Proposed Refactor

The new EdgeIndex object is a edge_index Tensor with lots of useful metadata. It is a drop in replacement to the usual edge_index Tensor.
The goal of this issue is to refactor pyg to utilise EdgeIndex across all modules. This should make our code more readable, and efficient in many cases.

• Updates to code where edge_index as sparse matrices are used. Instead leverage the EdgeIndex s matrix multiply operations.
• Updates to utils. Example sort_edge_index doesn't need to sort EdgeIndex if it is already sorted.
• Updates to transforms. Example update metadata like is_sorted when using transforms like AddSelfLoops.

[akihironitta]

@rusty1s I saw your comment in #8700:

EdgeIndex is not yet compatible with torch.compile

Does it mean that we should have this issue on hold for now?

[rusty1s]

No, it just needs some MessagePassing adjustments.