Related work: fix three stale cells, and cover all four families in the README

[salindne]

Documentation only — git diff --name-only master -- '*.mag' is empty.

The staleness

RELATED_WORK.md's comparison table still listed three of the six genus-3 ramified cells as unbuilt,
with *none* in our column and "Future work, PR6 / PR7 / PR8" beside them. All three were derived
and merged in those very PRs — the odd-characteristic doubling in #30, both characteristic-2 formulas
in #32. The arb doubling figure was stale too, quoted at 61 M+S / 3C / 92A where it has measured
58 / 4C / 80A since the PR36 sweep.

That is the worst kind of stale for this particular file, because it is the one a reader opens to find
out where the work stands against the literature — and it said we had no entry in three of the four
cells that have competition. In fact we are ahead on both axes in all four:

cell	previous best	ours	margin
odd char, addition	67 M+S, 105A† — Nyukai 2006	56 M+S, 0C, 59A	−11 M+S, −46 A
odd char, doubling	68 M+S, 93A† — Nyukai 2006	58 M+S, 0C, 61A	−10 M+S, −32 A
char 2, addition	67 M+S, 100A† — GKP 2004	54 M+S, 0C, 62A	−13 M+S, −38 A
char 2, doubling	69 M+S, 107A† — GKP 2004	55 M+S, 2C, 55A	−14 M+S, −52 A

The same staleness appeared in prose in both lane sections — "the doubling comparison still cannot be
made until nch2_ramifiedG3_DBL.mag exists" and "we have no entry in this lane yet" — and the
Birkner reconciliation was still written as something PR7/PR8 would do, where it was decided on
2026-08-24 to be a publication task rather than a repository one. All corrected.

The README gains a "Related work" section

The comparison against prior art is the first thing a reader wants, and it was previously reachable
only by opening a second file. It now sits in the README with the table above, and four caveats travel
with it rather than being left to be discovered:

• † marks A counts we derived, not figures the authors published — no prior work in this area
  reports additions at all.
• C is zero in every previous-best cell, because those normal forms leave no coefficient to
  multiply by. It is non-zero only for us and only in arb, which cannot normalise — so that row's 1C
  and 4C are the price of generality.
• Three of the four cells did not exist when this work started. Only the odd-characteristic
  addition was inherited.
• The char-2 rows are quoted at deg h = 3. Cheaper numbers exist at other h-shapes and are not
  comparable — Birkner's h = 1 doubling is 21 M+S, but a constant h collapses most of the work.

The harder target is quoted in every row: GKP is 70 M+S for both odd-characteristic operations against
Nyukai's 67 and 68, and their alternative char-2 variants are 68 and 72 against the 67 and 69 shown.

What moved, and what did not

Operation counts keeps what is about provenance — every row measured, and the 168 published cells
the counter reproduces — and hands the comparison material to the new section. The arb family's
comparison against the thesis's own split Degree-3 rows moved with it: 12 better on M+S, 11 on C and
16 on A for the addition, 18 / 15 / 21 for the doubling, which is what "ramified has strictly less to
do than split" predicts and which was not true when this work started.

Every margin was re-derived from the measured figures rather than carried across, and checked
arithmetically. The underlying counts come from verification/opcount.py, which executes the formulas
rather than reading a table.

Verification

Docs only, so the gates that apply are the static ones: check_paths --strict (259 references),
readme-paths (91 paths across 6 READMEs), and ruff. No formula, tester or tool is touched.

Update: all four families, not just genus-3 ramified

The section originally covered genus-3 ramified only, since that is the family the thesis defers. The
other three already have published comparisons — in the thesis itself — so the README was telling a
quarter of the story in the place a reader looks first. All four are now there:

family	compared against
genus 2, ramified	Lange 2005 (4-coordinate affine)
genus 2, split	Erickson, Jacobson and Stein 2011
genus 3, split	Rezai Rad et al. 2019, Sutherland 2019
genus 3, ramified	Nyukai 2006, GKP 2004 — compiled here

Every "This work" row was re-derived from verification/opcount.py and matches. Thirteen rows
across the three thesis comparison tables — five genus-2 ramified, four genus-3 split, four genus-2
split — every one exact. This is not redundant with PR41's check: the cost tables and the comparison
tables are separate transcriptions of the same quantities, and a transcription is where a copy error
hides. It found none.

One limitation stated rather than glossed: those three comparisons are as published in 2020, and no
search for work appearing since has been done. Only the genus-3 ramified survey is current, having been
compiled in 2026. So the three thesis rows are re-verified against our own measurements, not
independently re-surveyed — a weaker claim, and the section says so.