names: collapse X-bridge docs to a one-line note

[pudo]

Summary

The X-bridge case in run_cluster (a part referenced from two clusters when a later edge bridges two pre-existing clusters with no shared vertex) is structurally unreachable under the current monotone DP. The cursor walk in run_align can't produce the non-monotone overlap pattern an X-bridge requires. Confirmed empirically: 0 bridge events fired across the 818-case cases.csv corpus.

The plan docs and the run_cluster docstring previously framed it as a real-but-rare invariant violation that needed a union-find rewrite. After investigation that's misleading — the bug is real algorithmically but unreachable from real input today. This PR collapses the discussion to a one-line note in each of the three places it was tracked.

Background

Investigation triggered by a pudo/cluster-dsu branch (kept locally as a reference) that would have been a ~300-line rewrite — Dsu struct + iterative path-compressed find + union-by-rank + emit-order tick tracking + 6 unit tests. The rewrite is correct algorithmically and bit-identical on the corpus (0 outcome flips, 0 score diffs over 818 cases). But the X-bridge regression test it introduces exercises code that the real DP can never trigger — it builds synthetic AlignmentData directly.

The structural argument: align.overlaps is built by a single monotone DP walk through the SEP-joined strings. Both qry_idx and res_idx only advance, never backtrack. So if (qp_a, rp_b) accumulates an Equal step, no later step can be at (qp_c, rp_d) with qp_c > qp_a and rp_d < rp_b — but that's exactly what an X-bridge needs (two non-vertex-sharing edges, then a third that bridges them). Under sort-order processing on lex-sorted edges, the bridge edge always lands at a position where it's a chain-via-shared-vertex instead.

The DSU rewrite is the right shape if the connectivity-rule replacement lands (threshold = 0 expands which edges qualify and could unlock the case), or if the DP ever stops being monotone. Neither is imminent. Better to revert and pull the DSU back when there's a concrete reason.

What this PR changes

File	Change
rust/src/names/compare.rs (run_cluster docstring)	Replaces the "X-bridge limitation" paragraph with a short note explaining structural unreachability + what would need to change to unlock it.
plans/weighted-distance.md § Open spec knobs → Clustering rule fragility	Two sub-bullets collapsed to one sentence covering both threshold fragility and the DSU follow-up.
plans/arch-name-distance.md § Pairing rule	Drops the X-bridge limitation paragraph + redundant fragility bullets; keeps the threshold-fragility one-liner.

Net diff: 17 insertions, 55 deletions.

No code behaviour change. No test changes.

Test plan

• cargo test --release --features python — clean
• pytest tests/ — 470 pass
• mypy --strict rigour — clean
• cargo fmt --check, cargo clippy --all-targets -- -D warnings (with and without --features python) — clean

🤖 Generated with Claude Code