Refactor node attributes

[JPXKQX]

This PR introduces the NamedNodesAttributes(nn.Module) class, designed to enhance the management of node attributes within graph neural networks. Key features include:

• Node attribute tensor: The class provides a tensor that combines coordinates with trainable parameters for each set of nodes in the graph.
• Grid size: It retains the grid_size attribute, indicating the total number of nodes.
• Channel dimension: It tracks in_channels_src/dst, which represents the dimensionality of the node attribute tensor.

The main improvement is the ability to handle an arbitrary number of nodes. This development will facilitate extending AnemoiModelEncProcDec to other use cases, such as hierarchical models or multi-encoder/decoder.

• Tested on 1, 2 & 4 GPUs (with model sharding & transformer)
• Tested on 1 GPU: gnn, graphtransformer & transformer
• Tested on 1 GPU: lam & stretched

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.85%. Comparing base (deba13a) to head (9e40145).
Report is 2 commits behind head on develop.

Additional details and impacted files

@@           Coverage Diff            @@
##           develop      #64   +/-   ##
========================================
  Coverage    99.84%   99.85%           
========================================
  Files           23       23           
  Lines         1301     1350   +49     
========================================
+ Hits          1299     1348   +49     
  Misses           2        2           

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

[HCookie]

LGTM
Maybe consider adding some property typehinting

[JesperDramsch]

Thanks for tagging me in. Looks very nice. I have some questions, but I think generally this is also good to go.