refactor: decompose scan/extractSkills + typed SectionedResume (#137, #132)

[s-annam]

Summary

Two behavior-preserving refactors bundled onto one branch (epic run, disjoint blast radii):

• Reduce cognitive complexity of relocated extract/ extractors (fallow notes) #137 — decompose scan (extract/contact.ts) and extractSkills (extract/skills.ts) into smaller cohesive helpers to clear fallow's cognitive-complexity gate (>15). Logic preserved verbatim; snapshots byte-identical.
• Emit typed SectionedResume from cascade + retire skillsSectionText side-channel (proof PR) #132 — promote the section structure the cascade already computes into a typed SectionedResume on CascadeResult, and retire the skillsSectionText side-channel end to end. The scorer now derives the skills-exclusion set from sections.byName.get("skills"). Inert by design — corpus goldens unchanged.

Closes #137
Closes #132

Test plan

• npm run typecheck clean
• npm run test green (616/616, 41 files)
• npm run lint clean
• Zero *.expected.json churn (goldens NOT regenerated — clean diff is the pass condition)
• Manually verified in npm run dev / npm run preview

[rohithgollapalli]

Review — Approve ✅

Two clean, behavior-preserving refactors with disjoint blast radii. All gates pass locally.

Verification performed

• ✅ npm run typecheck clean
• ✅ npm run lint clean
• ✅ npm run test — 616/616 pass, 41 files
• ✅ Zero *.expected.json churn (git diff --stat origin/main...HEAD on fixtures is empty — the stated pass condition)
• ✅ No lingering code references to skillsSectionText/skillsSectionLines — remaining hits are all doc comments / docs/segmentation-spike.md

#137 — decompose scan / extractSkills

• contact.ts: extractLocation, extractOtherUrls, scan lifted out of the extractContact closure verbatim. Diffed line-by-line — location precedence (US→intl, @ guard), phone-region derivation order, LinkedIn fallback chain, portfolio/website bucketing all identical.
• skills.ts: tokenizeCell is a faithful extraction of the inner loop, preserving the critical ordering (drop contact-link cells before SKILL_SPLIT_RE shreds the URL).

#132 — typed SectionedResume, retire skillsSectionText

Behavior preservation hinges on byName.get("skills") being byte-identical to the old skillsSectionLines, and it holds:

• Old: findSection(sections, "skills") flattens all matching sections' lines in document order, then .map(trim).filter(nonEmpty). New: toSectionedResume iterates sections in document order and accumulates per-section trimmed/filtered lines. Per-line trim/filter ⇒ same array; accumulate-don't-overwrite correctly mirrors findSection's multi-match flatten.
• Scorer guard is equivalent: old if (!skillsSectionText) vs new if (!lines || lines.length === 0) — both collapse empty/absent to undefined. The split(/\r?\n/) round-trip is removed without changing the resulting set.
• The two no-Tier-1 defaults (runCascadeFromMarkdown no-markdown path, buildScannedResult) correctly yield byName.get("skills") === undefined, matching the previously-absent skillsSectionText.

Minor (non-blocking)

1. accomplishmentSections default inconsistency — toSectionedResume sets the canonical 3-name list, but the fallback defaults (cascade scanned / no-markdown paths and test helpers) set []. Inert today (documented as "not yet consumed"), but a future consumer reading it off a scanned result would get [] rather than the canonical list. Consider reusing ACCOMPLISHMENT_SECTION_NAMES in those defaults.
2. Cosmetic — extractContact now opens with a stray blank line (contact.ts:169) where the scan closure used to sit.
3. The extractContact doc-comment (contact.ts:39–55) is now ~125 lines above the function it documents (the refactor widened a pre-existing gap); optionally worth moving down onto the function.

None block merge.

🤖 Generated with Claude Code

[s-annam]

Addressed all three minor nits from the review in de869c0:

1. accomplishmentSections default inconsistency — exported ACCOMPLISHMENT_SECTION_NAMES from sections.ts and reused it in both cascade fallback defaults (no-markdown path cascade.ts:357, scanned-abandon cascade.ts:476) in place of []. A future consumer reading accomplishmentSections off a scanned result now gets the canonical list, matching toSectionedResume.
2. Stray blank line — removed the leading blank line in extractContact.
3. Orphaned doc-comment — moved the extractContact doc-comment back down onto the function (the scan/extractLocation/extractOtherUrls extraction had widened the gap).

Gates green on the branch: npm run typecheck clean, npm run lint clean, npm run test 616/616. Zero *.expected.json churn. Re-requesting review since the push dismissed the prior approval.