feat: add llms.txt support

[wellkilo]

PR: Add /llms.txt and /llms-full.txt (closes #34)

Closes #34.

What this PR does

Adds an llmstxt.org-compliant llms.txt (curated index) and an llms-full.txt (expanded one-shot context bundle) so LLM-powered tools — coding agents, IDE chat plugins, doc-aware search — can ingest the course without crawling the whole VitePress site.

Both files are authored at the repo root (the spec-required location) and copied into docs/public/ so VitePress serves them at the site root. A tiny generator script keeps the two copies in sync, and a CI workflow fails the build if they drift.

Files

Path	Kind	Purpose
llms.txt	new	Curated index, H1 + blockquote + sections per spec. Lists README, 12 lectures, 6 projects, harness-creator skill, resources, 13 locale READMEs (under ## Optional).
llms-full.txt	new	Long-form bundle: inlines repo structure, the five-subsystem definition, one-paragraph summaries for L01–L12, the project arc, the harness-creator reference card, and the canonical session lifecycle.
docs/public/llms.txt	new (generated)	Served at https://walkinglabs.github.io/learn-harness-engineering/llms.txt.
docs/public/llms-full.txt	new (generated)	Served at https://walkinglabs.github.io/learn-harness-engineering/llms-full.txt.
scripts/build-llms-txt.mjs	new	Sync script (Node built-ins only). --check mode used by CI.
package.json	edit	Add llms:build / llms:check scripts; chain llms:build into docs:build.
.github/workflows/llms-txt-check.yml	new	CI guard — prevents the two copies from drifting.
README.md	edit	Add a short "For AI assistants (llms.txt)" section after the Skills section.

Why this matters

• The repo is a high-visibility course (7.6k★) about how LLMs consume repositories — shipping an llms.txt is on-brand and lowers the friction for agents that want to reference the course mid-task.
• The spec is intentionally tiny (one markdown file, fixed shape). Cost-to-maintain is near zero, and the CI guard makes drift impossible.
• Putting the file in docs/public/ (not only at the repo root) ensures both github.com/.../llms.txt and the hosted site URL work — most agents check the site URL.

How I verified

• Ran node scripts/build-llms-txt.mjs locally — both files land in docs/public/ with matching byte counts.
• Re-ran with --check against a freshly synced tree — exit 0; mutated one file and re-ran — exit 1 with a clear drift message.
• Manually validated llms.txt against the llmstxt.org spec (H1, blockquote summary, sections of markdown lists, ## Optional section semantics).
• Locale READMEs and lecture/project paths cross-checked against the docs/.vitepress/config.mts sidebar tree.

Out of scope (intentionally deferred)

• Auto-generating llms-full.txt from lecture frontmatter. Doable, but adds a parser and risks summary drift. Current approach is a hand-curated bundle whose only sync requirement is "copy two files to docs/public/" — caught by CI.
• Per-locale llms.txt. The spec is silent on i18n; the English bundle already points at every locale README under ## Optional. Happy to add if maintainers want it.
• Inlining lectures 09 / 10. Tracked separately by 9,10两章没有内容 #2; llms-full.txt notes their tracked status.

Reviewer checklist

• npm run llms:build produces no diff after running it twice in a row.
• npm run docs:build still succeeds end-to-end (the new step is llms:build running before vitepress build docs).
• After deploy, the URLs below return 200 with Content-Type: text/plain or text/markdown:
  • https://walkinglabs.github.io/learn-harness-engineering/llms.txt
  • https://walkinglabs.github.io/learn-harness-engineering/llms-full.txt

[wellkilo]

@sanbuphy I have resolved issue #34 and would appreciate a code review. Thanks!