Add connectivity and degree distribution diagnostics for generated graphs

[Moka-4444]

Problem Statement:
Currently, when generating flexible graphs for irregular or sparse data, there is no built-in automated way to verify if the resulting graph is 'well-balanced' (e.g., detecting nodes with extremely high/low degrees or isolated components) before passing it to neural-lam.

Proposed Solution:
I propose adding a diagnostic utility (e.g., in visualise/diagnostics.py or as part of the validation layer) that automatically calculates and reports:

Node Degree Distribution: To ensure message passing remains stable across the grid.

Connectivity Check: Identifying isolated nodes or disconnected clusters that could lead to NaN losses during training.

Spatial Density Mapping: A visualization helper to see how graph density aligns with the input data grid.

Impact:
This will significantly help contributors (like myself) who are working on Project 1 (Flexible Graph Construction) to benchmark different strategies (kNN, Radius, etc.) and ensure they meet the 'well-balanced' criteria mentioned in the GSoC ideas.

I’m happy to start drafting a PR for this if the maintainers agree this is a useful direction!
@joel Oskarsson @hauke Schulz @leif Denby.

[Dhairya-45]

Thanks for opening this — I’d be interested in working on it.

My plan is to start with a small first PR focused on core diagnostics for generated graphs:

1. degree statistics / degree distribution summary
2. connectivity checks (isolated nodes, number of connected components, fully connected or not)
3. a lightweight interface that can be reused across different graph construction strategies

If that direction sounds good, I can begin with the analysis utilities first and keep visualization/spatial density mapping as a follow-up step so the initial PR stays focused and easier to review.

Please let me know if you’d prefer this to live under a diagnostics utility module or closer to the existing validation/visualisation code.

[Calculus0]

I can help with this.