Fix typos in comments and docstrings, and GIN script default model

[Avicennasis]

Summary

Fixes 8 typos in comments and docstrings, plus one copy-paste artifact in run_pgd_gin.py's argument defaults.

src/models/layers.py

• # Inspired and Adapted and inspired from ... -> # Inspired and Adapted from ... (duplicated "and inspired"; now matches the wording already used at the top of src/models/utils.py)

run_pgd.py

• # Set to seed -> # Set seed

run_pgd_gin.py

• # Let's init the GCN -> # Let's init the GIN (the line below it constructs GIN(...); the comment was carried over from run_pgd.py)
• ⚠️ Please review this one carefully - it is the only non-comment change: parser.add_argument('--model', ..., default='GCN') -> default='GIN'. This is the GIN driver script, so 'GCN' is a copy-paste leftover from run_pgd.py. It is a no-op at runtime today because args.model is never read in either script (the model class is hard-coded), but the default is shown in --help and would be wrong if the flag is ever wired up. Happy to drop this hunk if you would rather keep the change comment-only.

src/models/utils.py

• - Takes the number fold -> - Takes the fold number (both train_function and test_function docstrings - fixing only one would have left the two docstrings inconsistent)
• transform the adjacency matrices as block matrix -> into a block matrix (both docstrings; the code calls sp.block_diag(adj_batch), producing a single block-diagonal matrix per batch)
• it transform the -> it transforms the
• Sum of predictions with agree with labels -> predictions which agree with labels

No behavioural changes: everything except the default='GIN' line is a comment or docstring. Copyright/licence headers were left untouched.