feat: add SEO/GEO evidence workbook

[moose-lab]

Summary

• Adds a typed SEO/GEO evidence workbook that expands every registered Search/AI baseline row into a surface-specific audit task across Google Search Console, ChatGPT Search, Perplexity, Google AI Overview, Gemini, and Claude.
• Adds an editable 48-row CSV mirror for SEO/GEO supervisors to collect query, citation, accuracy, competitor, and forbidden-claim evidence without drifting from the active experiment contract.
• Adds Sites as a reviewable launch-package output: launch materials map channel, Launch site page spine artifact, Studio preset, full launch package scope, and qualification tests.

Functional Points

• buildSeoGeoAuditWorkbookRows() derives audit tasks from searchAiBaselineRows instead of duplicating experiment definitions.
• renderSeoGeoAuditWorkbook() produces a manual audit runbook that keeps the decision as insufficient data until GA4, Search Console, and AI-answer evidence exists.
• docs/marketing/data/seo-geo-audit-workbook.csv gives reviewers one row per baseline/surface pair.
• Site output is included in generated launch materials and artifacts as a page spine, not automatic publishing.
• Full launch package qualification can now capture sites scope while browser analytics only receives safe aggregate scope counts.

Verification

• npm test -- src/marketing/seo-geo-audit-workbook.test.ts src/marketing/search-ai-baseline.test.ts src/marketing/growth-experiment-report.test.ts
• npm test -- src/server/generation/generation.test.ts -t "source-backed free repo launch brief"
• npm test -- src/App.test.tsx -t "keeps the generator studio|full launch package contact|landing architecture"
• npm test -- src/server/marketing/launch-package-fit.test.ts src/server/marketing/lead-capture.test.ts -t "launch package fit|full launch package qualification"
• npm test -- --testTimeout=30000 --pool=threads
• npm run build
• git diff --check

Notes

• This is P0 measurement infrastructure and product-output enablement, not validated SEO/GEO demand.
• Default Vitest forks pool showed local worker startup timeouts; the full suite passed with --pool=threads.
• Public SEO/GEO, paid demand, pricing, citation, ranking, and revenue claims remain unvalidated until the evidence workbook and 14-day experiment windows are filled.