tools(mutation-teeth): a misnamed tooth says so, and the ledger places the seven rules no row claimed

[wefio]

Closes the two tails the derived-mutants arc left open, from main (818228ae).

1. tools(mutation-teeth): an expect that names no case says so

A tooth names the case that must fail, and the sweep filters a suite by that name so one tooth costs one case instead of a whole suite. When the name matches no case the filtered run passes with nothing in it, and that pass was reported as a surviving mutant - which is how two teeth were read as broken while they were only misnamed.

• ranACase (tools/mutation-anchor.ts, pure, no filesystem) reads the run's own report: a marked line that is not the suite file is a case; a suite line alone means the filter matched nothing; an unrecognized reporter answers unknown, so a tooth is never accused on evidence the harness cannot parse. The suite file's own line is what makes the answer possible - it runs whatever the filter did.
• The runner asks before it believes a pass, reports the tooth as misnamed with the name it wanted, counts it as a problem (non-zero exit), and skips the whole-suite fallback for it: the defect is the name, not the coverage.
• tests/tools/mutation-anchor.test.ts gained the case covering the file line alone, a matched case, a failed case, two suites where nothing matched, and the unknown reporter (35 cases in the file now).
• The record's own failure-mode note moves from "open" to answered, in the same commit.

2. docs(ledger): the seven rules no row claimed get one home each

The retirement pass left seven teeth whose rules no row named. Each rule was recovered from the register before the retirement, each replacement case was located and its assertion read, and the answer is not seven new rows:

tooth	its rule's home
round-publication-opens-its-own-transaction	B3 - the case is in B3's suite and asserts the transition's rollback for the publication half
round-never-releases-its-pin	B4 - the other side of retention: the pins go with the values they protected
ordered-mode-becomes-any-topological-order	A6, a new row - the design's sentence about publication order had no row (A: 5 -> 6 rows)
the-loop-awaits-each-unit-instead-of-the-batch, a-unit-is-dispatched-twice-in-one-batch	F2b-slot - both share one case, now named there
a-unit-nothing-checks-is-still-a-unit, the-parent-check-ignores-its-own-verdict	F2c - the two cases now named there

The readings line no longer calls them orphans, and the record no longer leaves the question open: what remains open is only whether "a check but no row" is a defect at all - it was one here, but the defect was in the ledger, not in the register.

Verification

• whole-register sweep, with the misnamed check live: mutants: 110 of 110 caught by the named test, all 110 by the case each expect names, 22 of 22 targets restored byte-identically, exit 0 in 162 s - i.e. the check raises no false alarm anywhere in the register
• a deliberately misnamed tooth reads misnamed (no case in this target's suites is named "..."), the file is restored byte-identically, exit non-zero
• npm run verify:static exit 0; anchors: 110 of 110 resolve, over 22 targets; docs:check 304 files, 0 errors, 0 warnings; lint 0 findings; complexity gate ok
• npm run test:product 1581 pass, 0 fail, exit 0 (1580 + the new case)
• no product behaviour changes: a research tool, its tests, and the documents that describe them