Feature request: Control aspect ratio

[ewels]

It'd be nice if we could control the aspect ratio of the output SVG. For example, make it thin and wide.

[pinin4fjords]

Parking this for now. The feature itself works (PR #666, paused): it picks the fold_threshold whose rendered shape is closest to the target and reports honestly what it achieved. The blocker is downstream of the feature - the layout engine doesn't yet produce good maps in the regime that "tall" aspect ratios push it into: low fold_threshold, sections reoriented TB, serpentine multi-row grids. nf-metro render examples/rnaseq_auto.mmd --aspect 0.5 lands squarely in that regime and the result is poor (#670, #671, #672, #664).

So this is blocked on layout-engine quality, not on the feature. The feature is the right messenger and is ready to un-pause once the engine can handle the flexibility it asks for.

Readiness criteria (when to revisit)

Re-test with nf-metro render examples/rnaseq_auto.mmd --aspect 0.5, plus one or two other fold-driven pipelines (variantprioritization). The engine is ready when a "tall" render is as clean and legible as the default LR render:

• Entry top-left, rows left-aligned. The source section anchors top-left; serpentine return rows share the left margin; no large empty quadrants. (Aggressive auto-fold strands the entry section off the top-left and wastes canvas #670)
• TB sections lay out cleanly. A section reoriented top-to-bottom by folding lays its fork/join fan out without weaving or overlapping on the vertical runs. (TB-reoriented stacked sections weave their internal fan lines #671)
• Clean fan-in in small sections. Two-producers-to-one-sink inside a tight section drops straight / nests rather than bowing into a crossing tangle. (Small fan-in sections route converging lines as a crossing tangle #672)
• Concentric corners nest. Section-exit down-turn bundles hold constant radial spacing through the bend. (Uneven concentric spacing at the down-turn out of section 1 (genomeassembly) #664)
• No overlaps, no validator regressions in the tall render (tests/layout_validator.py clean).
• Meaningful granularity. Fold-driven pipelines expose more than two achievable aspect ratios, so the target snaps to a useful shape rather than just the tall/wide extremes. (rnaseq currently only reaches 2.03 and 3.53.)

This overlaps with #533 (genuine TB primary axis) and #548 (declarative concentric radii); progress there ticks several boxes above.

When all of these hold on rnaseq plus one or two other fold-driven examples, un-pause #666.

[pinin4fjords]

Data point from authoring a dense map (nf-core/riboseq, ~30 station-columns
across 8 sections) on main @ 5104665: the one lever that looks like it
should help here - fold_threshold - only works in the "unfold" direction.

--fold-threshold above the default does what the docs say (it keeps a
long trunk on one row):

examples/rnaseq_auto.mmd   ft=6 -> 1509x585   ft=10 -> 1509x585
                           ft=15 -> 1509x585  ft=25 -> 2135x435

but lowering it never folds earlier - 6, 10 and 15 are byte-identical
canvases. On examples/variantbenchmarking_auto.mmd and on the riboseq map
(3268x916, ratio 3.6 at default) every value from 6 to 25 gives the identical
canvas, so there is no way to ask for an earlier wrap.

The practical effect is that the only way to control aspect ratio on a wide map
today is hand-authored %%metro grid: cells - and because grid columns are
shared between rows, an explicit grid tends to introduce dead space unless you
use the packed-cell form (%%metro grid: a, b, c | 0,0). If fold_threshold
worked downward it would cover most of what this issue asks for.