Optimize node_swap

[maximelucas]

Optimize node_swap by replacing the three-loop temp-ID swap with a single-pass dictionary mapping. Closes #642

Old approach: used a temporary node ID to swap in three sequential passes — replace nid1 → temp, then nid2 → nid1, then temp → nid2 — plus a deepcopy of the full edge dict.

New approach: builds a {nid1: nid2, nid2: nid1} mapping and remaps each edge's members in a single dict comprehension. Handles the edge case where both nodes appear in the same edge correctly (the swap is applied
atomically per member, leaving the set unchanged). Removes the id_temp parameter and the deepcopy import entirely.

Timing

Single swap on an xgi.random_hypergraph averaged over runs.

nodes	edges	before (ms)	after (ms)	speedup
50	324	3.13	2.08	1.51x
200	7 533	70.69	46.71	1.51x
500	24 396	248.70	160.23	1.55x
1 000	93 195	958.04	643.82	1.49x

The ~1.5x speedup is consistent across sizes.

Tests

Added some tests for uncovered cases.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[leotrs]

Nice cleanup Max. The dict-mapping comprehension is much easier to read than the three-pass temp-ID swap, the deepcopy goes away, and the new tests cover the previously implicit edge cases (self-inverse on both-in-one-edge, original-not-mutated, attribute preservation).

One thing worth flagging: removing the id_temp parameter is a backward-incompatible change. Anyone passing node_swap(H, a, b, id_temp=...) will get a TypeError. The arg was always implementation-detail leakage so dropping it is the right call, but worth mentioning in the changelog so it's an intentional v1.0 break rather than a silent one.