v3.70.6

Land the hand-written spacing on the v2 ladder (#205)

ui/src wrote 23 arbitrary spacing values by hand across fourteen
distinct numbers — gap-[7/9/10px], px-[9/10/13px], py-[1/6/7px],
pt-[2/10/11px], pb-[3px], mt-[7/9px], inset-[2px], inset-x-[1px],
inset-y-[4px]. spacingRoles.test.js counted them and stopped the number
growing, and said outright that landing them on a ladder was the real
answer and needed a mockup set and the owner naming one.

Both happened. .mockups/spacing-scale/all.html rendered four directions
against a v0 reference and v2 was chosen: ladder 0/4/8/16, sub-4 nudges
deleted rather than rounded, and no 12. The mapping is 1,2,3 -> 0;
4 -> 4; 6,7,9,10,11 -> 8; 13 -> 16.

All 23 sites are now Tailwind scale utilities. The rem base is 16px —
the 14px font-size is on body, not html — so gap-2 is 8px, px-4 is 16px
and inset-y-1 is 4px.

--sp-ledger-inset goes 11px -> 8px with them. It was extracted at its
rendering value on purpose, because naming a value and moving it are two
different jobs and docs/SCREENS.md puts the second behind a variant set
and a pick. This is that pick. 8px rather than 12px because 12 is not on
v2's ladder and everything from 6 to 11 lands on 8.

That move is safe in the direction it went: shrinking the inset gives
each ledger cell 6px more room inside a LANE_COLS track whose width did
not change, so nothing can overflow. Growing it is what would squeeze
contents out of a column sized for them, and a future move upward has to
bring LANE_COLS with it.

CENSUS is now empty, which makes ARBITRARY_BUDGET 0. The ratchet fails
on the FIRST new arbitrary spacing value rather than the twenty-fourth.
That tightening is deliberate: there is no longer a backlog for a new one
to hide in, and a genuine one-off still gets in by adding its line with
its reason.

One guarantee got WEAKER as a side effect and is restored explicitly.
At 11px the ledger token's value was distinctive, so a cell that drifted
off it had to write px-[11px] and the ratchet caught it. At 8px the value
collides with px-2, so a drifted cell renders pixel-identically through
an ordinary utility that nothing would flag — and the existing four-use
count does not cover it either, since a FIFTH cell written with px-2
leaves that count at four. spacingRoles.test.js now asserts every
ledger-cell class string also carries px-[var(--sp-ledger-inset)], with a
fixture proving the check fails against a cell written with px-2.

tests/dossier-page.spec.ts already reads the token at runtime and asserts
every .ledger-cell paints exactly it, so the ledger alignment was guarded
at 8px without changing that spec at all.

Co-authored-by: NOC Dashboard <noc-dashboard@users.noreply.github.com>