fix(cli): support Chinese search and cold-start bootstrap

[Disdjj]

Summary

• segment Chinese natural-language queries with Intl.Segmenter, preserve exact-phrase ranking, and use an annotated CJK-bigram fallback only after zero lexical recall
• let new create the first llmdoc/ directory at the nearest Git root and provide actionable bootstrap guidance
• distinguish unborn HEAD in init-state, improve meta.missing remediation, and align the bootstrap order with validate -> commit --all
• add regression coverage and synchronize CLI schemas, design docs, website docs, and tracked llmdoc knowledge
• prune 475 tokens of reconstructable llmdoc detail while preserving all eight canonical owners and routing

Verification

• npm test — 64 tests passed
• npm --prefix cli run lint
• npm --prefix cli run typecheck
• npm --prefix website run build
• npx -y @tokenroll/llmdoc validate
• routing probes for all newly mapped implementation and test files report zero unmapped files
• post-prune report: ~2203 tokens, below growth gate, no duplicate or fragment candidates

Closes #59