Linguistic competence metrics + counts-only vocabulary dump

[willwade]

Summary

Two related capabilities for longitudinal AAC metrics work (Grid-Metrics-Exporter consumes both):

1. Linguistic competence metrics (privacy-preserving)

• Source- and language-agnostic core: MATTR-30 lexical diversity, syntactic (injected closed-class set), morphological (injected inflection classifier), spelling validity (optional dictionary)
• No word lists in the core — language resources are injected; missing resources report unavailable with a reason, never silently degrade

2. Counts-only vocabulary dump

• dumpVocabulary(tree, { metrics }): inventories vocabulary by source — labelled buttons, Grid 3 page wordlists (<WordList>), prediction dictionaries (Prediction.PredictThis), and smart-grammar word forms — with per-source POS breakdowns
• Emits counts only, never word lists — safe for privacy-preserving reports
• PagesetSummary.vocabulary field so competence reports can carry it
• CLI: aac-processors vocabulary <file> (JSON out, --out, --no-smart-grammar)
• Declares AACPage.wordListItems on the interface (was runtime-only)

Testing

• New: test/vocabularyDump.test.ts (9 tests, incl. real example.gridset asset) + CLI integration test
• Full suite: 90 suites / 808 tests pass; npm run lint and npm run type-check clean

Notes

• Grid-Metrics-Exporter currently vendors a mirror (against pinned 0.3.2) and will switch to this API in its next versioned bump.