Add lexical richness indices (Brunet's W, Honoré's R) to competence timeline

[willwade]

Summary

Adds total-sample lexical-richness indices to the linguistic-competence
timeline: Brunet's W and Honoré's R, plus the hapax-legomena count.

Why

MATTR-30 (the headline lexical-diversity measure) depends on a moving window
and behaves awkwardly on the small, highly variable samples typical of AAC
output. Brunet's W (N·V^-0.165, lower = richer) and Honoré's R
(100·ln N / (1 − V₁/V), higher = richer) are total-sample type-token
measures that complement it without window-size sensitivity. Both are used in
DEPAC's lexical-complexity feature set (Tasnim et al., DEPAC 2022) and the
wider clinical-linguistics literature.

What

• lexicalRichness(words) — pure function of the token-frequency
  distribution; language-agnostic, no word lists or resources needed (fits
  the data-free competence core). Returns { brunetsW, honoresR, hapax, types, tokens } with null guards (empty stream; R undefined when every
  type is a hapax or N ≤ 1).
• MonthBin.lexicalRichness — computed per calendar month in
  analyzeTimeline (nulled on suppressed months alongside the other measures).
• Only aggregate counts/ratios emitted — no word lists (privacy unchanged).

Testing

• 5 new unit tests: closed-form values for a known distribution, null guards,
  monotonicity (W decreases as vocabulary diversifies at fixed N), and
  MonthBin wiring in analyzeTimeline.
• Full suite: 91 suites / 841 tests pass; lint (0 errors) and type-check clean.

Consumers

Grid-Metrics-Exporter dashboard already renders lexicalRichness charts from
the timeline (omitted gracefully on ≤ 0.3.3 reports); it will pick this up on
the next versioned release.