Fix entry-port fork fanout legs collapsing onto one opening vertex

[Copilot]

Entry-port-originated same-line fanouts in LR/RL sections could open from a shared source-side turn, producing a starburst instead of a depth-ordered staircase. The affected path was split across plain intra-section diagonals and entry-runway-owned branches, so only some siblings peeled correctly.

• Routing

  • Broaden the fused-opening-leg seating predicate to include flow-side LR/RL entry ports when the router, not a separate seating mechanism, owns the opening geometry.
  • Apply the same peel-delay budget in _route_entry_runway() that _route_diagonal() already uses, so deeper runway-owned siblings peel earlier instead of reusing the same lead X.

• Regression lock

  • Add a focused regression test covering:
    • the reported riboseq_inter_row_corridor repro
    • a minimal synthetic same-line entry-port fan
  • The assertion is mechanical: for same-direction siblings, deeper targets must start their opening earlier than shallower ones.

• Shape of the fix

  • Reuse the existing fused-opening mechanism rather than introducing a second port-only seating path.
  • Keep the change scoped to flow-side LR/RL entry-port forks; other port cases remain on their existing routing paths.

def _source_needs_fused_opening_seat(edge, src, ctx) -> bool:
    if not src.is_port:
        return _fan_plan_seats_the_fork(ctx, edge.source)

    port = ctx.graph.ports.get(edge.source)
    section = ctx.graph.sections.get(port.section_id) if port is not None else None
    return (
        port is not None
        and port.is_entry
        and section is not None
        and (
            (section.direction == "LR" and port.side is PortSide.LEFT)
            or (section.direction == "RL" and port.side is PortSide.RIGHT)
        )
    )

[github-actions]

Render preview is ready for review:
🖼️ https://seqeralabs.github.io/nf-metro/_pr/1965/

This preview shows only the renders that changed compared to main.

[pinin4fjords]

Closing — the underlying issue (#1831) has been closed as won't-do. On review, the pre-fix starburst rendering doesn't read as a problem on the real map it was reported from; the staggered version isn't a visible improvement for single-line forks. This change also introduced a genuine new line crossing in examples/topologies/funcprofiler_upstream.mmd (failing test_no_bridges_without_crossings in CI), unrelated to entry ports, so it isn't mergeable as-is regardless.