fix(graphs): keep edge_weights/node_weights/coords in sync after add_edges_from

[agu2347]

What

BaseGraph._reset_dicts() seeds _edge_weights (and _node_weights) from the graph's current nodes/edges, but is only ever called at construction time (or explicitly, by the alternative constructors). add_edges_from is inherited directly from nx.Graph and never routes through it, so edges (and any nodes they implicitly create) added after construction are silently missing from edge_weights, with no error or warning:

from qoolqit.graphs import DataGraph

g = DataGraph.from_nodes([0, 1, 2])
g.add_edges_from([(0, 1), (1, 2)])

print(g.edge_weights)  # {} -- expected {(0, 1): None, (1, 2): None}

This repo's own test suite was already routing around the bug -- test_to_matrix_unweighted called graph._reset_dicts() right after add_edges_from with a # FIXME: ... see issue #431 comment above it.

Fix

Override add_edges_from on BaseGraph (inherited by DataGraph, which doesn't define its own): call through to nx.Graph.add_edges_from, then seed any node/edge that's genuinely new with a weight/coordinate of None, matching _reset_dicts()'s own convention. Existing weights are left untouched, so this can't clobber edge_weights/node_weights set explicitly after construction.

New nodes are also handled (not just edges), since add_edges_from can implicitly introduce nodes that don't exist yet -- leaving those out of _node_weights/_coords would surface later as a KeyError in unrelated code paths (e.g. to_matrix(), distances()).

Testing

• Removed the _reset_dicts() workaround (and its # FIXME) from test_to_matrix_unweighted.
• Added test_add_edges_from_updates_weights_and_coords, covering: the exact case from the issue, that existing weights survive a later add_edges_from call, and that edges introducing brand-new nodes extend node_weights/coords for those nodes too.
• Ran the full test_base_graph.py suite (47 tests): all pass except one pre-existing, unrelated failure (test_from_nx_wrong_pos_attr, a & vs and short-circuit bug in from_nx's validation, confirmed present on main before this change too).
• Reverting just the base_graph.py change makes the new regression test fail with the exact symptom from the issue (edge_weights == {} instead of {(0, 1): None, (1, 2): None}).
• ruff check / ruff format --check clean on both changed files.

Fixes #431

[sgrava]

Hi @agu2347 ,
Thanks for proposing the bug fix! We really appreciate external inputs.
Unfortunately we were already fixing this in a separate PR #438 .

If you wish to contribute to the library, please consider other open issues.
Thanks.