feat: judge every build with one yardstick

The workbench and the maturity report each carried their own copy of
the same measurements, and the registry's ordering rule lived in three
places at once, so "how good is this build" could quietly mean
different things in different rooms. internal/measure now defines the
axes once, and both tools become renderers of it.

The yardstick is aligned with the conventions the payloads speak: a
text category is whatever atlas.render.as says it is, the same reading
the viewer makes, so the metric and the behavior can never drift
apart. New axes measure adoption of the vocabulary itself -- how much
of it a build declares, whether it says anything the registry does not
know -- and a build from before the conventions says so plainly rather
than scoring zero on a test it never sat.

commit

aad407a005521053b4da56ad5220f5610c415aa1

parent

3477b5c020e255ab0fd015ce5574c217532ad379

author

Dami <dami.etoile@gmail.com>