Add a marketing landing page at the GitHub Pages root

[coddingtonbear]

What this does

Replaces the typedoc output at the root of https://coddingtonbear.github.io/markdown-patch/ with a marketing landing page, and moves the API docs under /api/. The design was chosen from three reviewed mockups (docs-as-landing archetype with an interactive demo hero, plus a CLI read/write section and a "built for agents" section).

• site/index.html — the whole site: one hand-written, dependency-free static file (no generator, no external requests, light/dark via prefers-color-scheme, prefers-reduced-motion respected).
• .github/workflows/docs.yml — the build now assembles _site/ from site/ (root) + typedoc output (/api/) and uploads that instead of docs/ directly.
• README.md / package.json — the API-docs link now points at /api/; homepage points at the site.

Evidence this adds no risk

• The interactive demo cannot lie: all six demo results (append, marker rename, frontmatter replace, within list append, readTarget, projectMap) are precomputed, and each one was verified byte-for-byte against the real engine built from this branch — including the real version token 2b0a04 for the demo document. Verification script output: 6/6 PASS.
• In verifying, a README drift was caught, not copied: README's "Inspecting a document" example shows headings as an array of paths, but the engine actually returns a nested object ({"Weekly Sync": {"Notes": {}, ...}}). The site shows the engine's real output; the README fix is tracked separately.
• No engine code changed: src/ is untouched; the full suite passes (20 suites, 511 tests).
• The workflow's assembly step was executed locally (cp -R site/. _site/ && cp -R docs/. _site/api/) and produces the expected layout: index.html at root, typedoc tree under api/.

Known trade-offs / honest gaps

• Old API-doc URLs break: everything previously at /<page> (typedoc's classes/…, interfaces/…) moves to /api/<page>. No redirects are added; anything deep-linking the old paths (including search-engine results) will 404 until re-crawled.
• The demo is precomputed, not live — results are engine-verified but static. Bundling the actual engine for the browser is a possible follow-up.
• The token-count bar chart is labeled illustrative on the page, and is.
• Page rendering was reviewed as a served artifact during mockup review, but not re-screenshotted post-conversion to a standalone file; the conversion changed only the document shell (doctype/head) and link targets.
• The Pages deploy itself only runs on merge to main — the workflow change is exercised for real the first time then. Its steps are plain cp, run locally above.

🤖 Generated with Claude Code

https://claude.ai/code/session_01VbjyyWq1fcRe3iBg69oxtX

[coddingtonbear]

Verification run on this branch:

Unit suite (the project's full test surface — no separate integration suite exists):

Test Suites: 20 passed, 20 total
Tests:       511 passed, 511 total

Demo-content verification (every result the landing page displays, checked against the engine built from this branch):

PASS  append to Notes
PASS  rename Attendees -> People
PASS  frontmatter status -> final
PASS  within -1 append '- Priya'
PASS  readTarget Notes
PASS  map output
ALL_DEMO_CHECKS_PASS

Not exercised by these runs: the Pages workflow end-to-end (only runs on push to main; its cp-based assembly step was executed locally and produces index.html at root + typedoc under api/), and a browser render of the final standalone file (reviewed as a served artifact pre-conversion; the conversion touched only the document shell and link targets).