Auto-folded serpentine: return-row inter-section connector slopes instead of running level

[pinin4fjords]

Summary

In an auto-folded serpentine (no grid/direction directives - auto-layout wraps the section chain into two rows via a vertical turn-around section), the inter-section connector between the two sections in the return row is not level: it runs on a slope instead of a horizontal axis-aligned segment.

This is a cosmetic defect on a render that succeeds (no abort) - the auto-fold is exactly the layout we want, it just needs the connector to be horizontal.

Repro

assets/metro_map.mmd from nf-core/rnadnavar#83 (commit 0b11241f):

nf-metro render assets/metro_map.mmd --theme nfcore --fold-threshold 15 -o map.svg

Auto-layout folds it: row 0 is Pre-processing → GATK → Variant Calling → Normalization & Annotation (the last is a vertical TB section on the right edge that turns the flow around), row 1 is Consensus & Filtering → Realignment → Reporting flowing right-to-left.

The connector ensemblvep -->|dna,rna| vcf2maf (section 4 Normalization & Annotation → section 5 Consensus & Filtering) leaves the bottom of section 4, but the horizontal run into vcf2maf is tilted: the bundle exits at one Y and vcf2maf's entry sits a few px lower, so the dna+rna bundle slopes down-left into the station rather than arriving level. The two lines are also slightly non-parallel approaching the station.

Expected

The return-row inter-section run should be horizontal (axis-aligned), the same as the top-row runs - the exit Y off section 4 should match the entry-station Y in section 5 (or the connector should add a short jog and then run level), not a continuous slope.

Notes

• Pure auto-layout, default --fold-threshold 15, nf-metro 1.0.0. No directives steer this.
• Sibling defect in the same render: the realignment/rna lines cross at the section-5 exit (filed separately).

[pinin4fjords]

Sibling defect in the same auto-folded render: #1199 (peel-off crossover at the section-5 exit). Both surfaced on nf-core/rnadnavar#83 @ 0b11241f.

[pinin4fjords]

The #1199 fix (PR #1203, merged) added this exact render to the gallery as examples/topologies/foldback_exit_peeloff.mmd (the rnadnavar map, logo line dropped, fold_threshold: 15 pinned), so this slope now reproduces in-repo without the external file:

nf-metro render examples/topologies/foldback_exit_peeloff.mmd --theme nfcore -o map.svg

Measured mechanism for the normalization -> consensus connector (section 4 TB LEFT exit -> section 5 RL RIGHT entry): both ports sit at y=350, but their per-line offsets are misaligned by a constant +4px:

• normalization__exit_left_4: dna=4, rna=0
• consensus__entry_right_10: dna=8, rna=4

So each line steps +4px across the 52px gap and both connector segments slope by ~0.077. The bundle order is preserved (dna below rna on both sides); it's purely that the reversed consensus entry bundle is anchored 4px lower than the TB exit bundle. The fix is to reconcile the TB-exit and RL-entry per-line offsets across this fold seam (the exit/entry offset passes in routing/offsets.py), not a routing-time jog.

Regression infra is already in place from PR #1203: foldback_exit_peeloff is xfailed in test_topology_validation::test_no_almost_horizontal_edges referencing this issue, and carries a documented serpentine-family seam residual. When this is fixed, flip that xfail to a positive assertion so the level connector is locked.