ledger: features are first-class citizens

The merge ledger spoke only of pins: donorPins counted the offering, matched
and added and held accounted for it, and a donor's paths and areas fell out
of the account entirely -- mergeWorld skipped every shape collection without
a word. The account now covers every kind. Each mergedSource carries
donorFeatures, the donor's whole offering counted per kind, with donorPins
kept as the older spelling of its point count until flag day; addedShapes
holds the ledger's place at zero for the day a shape learns to merge. Origin
accounts fill the counts from the world's own collections, so a composed map
says up front how many points, paths, and areas it opened with.

What cannot merge yet is held on the record instead of dropped: mergeWorld
writes one held line per donor shape feature, title and all, under the one
reason 'shape features do not merge yet'. Matching itself stays point-only
and the alignment still stands on point anchors alone -- nothing about the
resolution changed, only the honesty of its account.

mergeGate audits per kind. The two point spellings must agree, the point
arithmetic must close over matched, added, held points, and rejected, the
held shapes must equal the donor's paths and areas, and the reserved
addedShapes must still be zero. The identifier walk now covers features of
every kind -- shape and point ids share one space -- and the winner's
per-kind counts must equal what its ledger sums to. No source pair produces
a shape-carrying donor today (only arcgis emits regions, and it never merges
into another source), so the held-shapes path is proven by fixture worlds in
tests rather than by the corpus.

internal/measure reads donorFeatures when present and lets donorPins stand
in for bundles from before the field existed; cartograph's volume table and
the maturity report print the donor's shape count beside its pins. The
Westminster bundle regenerates with exactly one payload change: the origin
account gains its donorFeatures line.

commit

13e21abda9e1f03ca8751d2a159e95155454ca0e

parent

d39f89df01c363626d10f6bca4795da4834a5c2f

author

Dami <dami.etoile@gmail.com>

kind

feat risk score 1.9 of 3, Jev's estimate