feat(layout): write line caches for bold characters in 한양신명조 and 한양중고딕

[airmang]

Summary

• write_line_caches left every paragraph holding bold characters in a 한양 face for Hancom Office, because FormFit laid those glyphs out at their regular advance.
• Since fix(form_fit): lay bold glyphs in the 한양 faces out a twentieth of the em wider #501, FormFit lays bold glyphs in the 한양 faces out a twentieth of the em wider, as Hancom does. Paragraphs with bold characters in 한양신명조 and 한양중고딕 now get caches.
• Bold characters in 한양견고딕 and HCI Hollyhock are still left for Hancom: some of their bold glyphs take other advances.

Test plan

• The look tests now expect caches for bold 한양신명조 and none for bold 한양견고딕 or HCI Hollyhock; the first fails before the change
• Line cache tests, lint, format, type checks and the other repository gates

🤖 Generated with Claude Code

[airmang]

Reviewed exact head 5e4002c, _bold_sure/_sure_of, and FormFit's bold advance path. All current CI checks pass; verified workflow head and inspected logs (6,929 passed, 18 skipped, 1 xfailed, lint/types/hygiene pass). 156 local cache/FormFit tests passed. Holding for the uncovered non-100% Hanyang-bold eligibility case described inline. The reproduced behavior establishes that an unverified metric path is now accepted for authoritative cache writing; actual Hancom width at those ratios has not been verified in this review. COMMENT because this account owns the PR.

[airmang]

[P2] Restrict the new bold opt-in to ratios FormFit has verified. Removing these two faces from the exclusion enables them across _SURE_RATIOS=50..200, while measure._laid_out only applies the Hanyang bold widening inside ratio == 100 (as #501 explicitly documents). On this head, a 14pt 한양신명조 bold paragraph at ratio=90 writes a cache (written=1), but char_advance('가',14,...) is 1260 for both regular and bold; ratio=110 similarly gives 1540 for both. The added test covers only default ratio=100. Since a cache fixes line breaks in Hancom, this should remain excluded outside the measured bold ratio until saved-Hancom fixtures verify those combinations. Please guard the applicable script ratios for these bold faces, or extend/verify their metric path before admitting them.

[airmang]

Re-reviewed exact new head 98356fc. Range-diff against the previously reviewed head shows rebase/CHANGELOG placement only; the bold eligibility and tests are unchanged. All 53 local line-cache tests and all 13 exact-head CI checks pass; workflow head verified and logs inspected (6,930 passed, 18 skipped, 1 xfailed, lint/types/hygiene pass). Existing unresolved inline finding PRRT_kwDOPxXhd86tiXjd remains applicable: bold Hanyang caches are admitted outside ratio=100 although that widening path is measured only at 100%. Hold remains; no duplicate inline finding.

[airmang]

Re-reviewed exact new head 3d5a047 and surrounding code. The updated head retains the previously reviewed implementation while incorporating the latest base and fixture/changelog merges. The existing unresolved Hanyang-bold ratio finding remains: ratio 90/110 cache writes are admitted while the bold advance equals the regular advance outside ratio=100. Exact-head reproduction still writes those caches; no new inline duplicate.

All 55 related local tests passed. All 13 exact-head CI checks pass; verified workflow head SHA and inspected logs (6,933 passed, 18 skipped, 1 xfailed plus lint/type/hygiene gates). Passing checks do not cover the reproduced issue; hold remains. COMMENT because this account owns the PR.

[airmang]

Re-reviewed exact new head 6008ce8 and its range-diff against the prior review. Changes are base/CHANGELOG or fixture-list integration; the implementation underlying the existing unresolved inline finding is unchanged.

Hanyang-bold eligibility at ratios 90/110 still writes authoritative caches while bold and regular advances are identical outside the measured ratio-100 path. Hold remains; no duplicate inline thread added.

All 56 related local tests pass. All 13 exact-head CI checks pass; workflow head SHA verified and test/lint/type/hygiene logs inspected (6,938 passed, 18 skipped, 1 xfailed). Those checks do not cover the reproduced finding. COMMENT because this account owns the PR.