README: lead with the operation counts, and say what is missing rather than what passes

[salindne]

529 lines to 491, and the reordering matters more than the length.

The counts lead

They are the result of this project, and they sat at line 70 — below a gate scoreboard, a per-family
coverage table and a four-branch exemption note. They are now the first section, with the genus-3
ramified rows grouped beside their genus-2 counterparts rather than split across the table by model.

The source column is given the emphasis it needs. Measured means verification/opcount.py
executed the formulas and identified the frequent case by observing which branch was taken.
Published means read from the thesis, because opcount.py cannot measure split families and
refuses rather than guessing. Half the table is therefore not independently checked by this
repository, and a reader skimming for numbers has to be told that in the same breath as the numbers.

The six-cell matrix is gone

It was a progress tracker: its content was the word "missing" in three cells. Every cell ships now, so
it says only that a thing is finished — which the counts already imply. One sentence replaces the grid.

Status becomes "Current and planned work"

The old section was ~93 of the file's 529 lines and was mostly evidence: five gate figures, coverage
by family, the measured-versus-published note, an operation-count table, a not-runnable line. Evidence
belongs beside the commands that reproduce it, so the gate figures moved into Testing, each with
its command in its own column.

What replaces it is what is missing: the Birkner/GKP comparison the char-2 formulas owe, the
split-family blindness in opcount.py and the two proved savings waiting on it, E20's
corpus-adequacy gap, the absent positive-reduced basis at genus 3, and the dead converter.

That section absorbed Known gaps and roadmap, which was saying the same thing in another place —
and which was wrong. Its first bullet read "Genus 3 ramified is arb and nch2 only. The ch2
specialisation is still to be derived, so five of the six cells exist." PR7+8 (#32) closed that and
the bullet was missed. A README that states a shipped cell is missing is worse than a long one.

Every figure re-measured, not copied

This file has carried stale numbers three times in this series — tester counts, corpus size, the
matrix row — each time because a figure was moved rather than re-derived. So all of them were re-run:

• whitebox 1,886/1,886, coverage 1,925/1,929
• driver --strict 13,746 compared, 0 wrong
• selftest 17 sections
• the suite's 30 testers, counted from its own per-tester logs rather than from run_family calls
  (17 calls is not 34 testers — three are skip_test paths)
• all twelve op-count rows re-run through opcount.py, including the six genus-2 rows that were
  not being changed, since the point is that a moved figure is a re-derived one

Also

Two links to the deleted #status anchor were repointed at Testing rather than left dangling, and one
bare opcount.py became the real path so readme-paths resolves it.

Gates: Magma suite 30 passed, 0 failed, 0 skipped; whitebox 1,886/1,886; selftest 17/0;
readme-paths and check_paths --strict both clean.