Support final (sofit) Hebrew letters in gematriyaStrToNum

[mjradwin]

Problem

gematriyaStrToNum silently ignored the five final (sofit) letters ך ם ן ף ץ, which dropped the tens digit of any Hebrew year ending in 20, 40, 50, 80, or 90:

gematriyaStrToNum('תש״ף'); // 700 — expected 780

Since a Hebrew year's last letter falls in final form, every such year collapsed onto its bare century. This surfaced through HDate.fromGematriyaString('א׳ שבט תש״ף'), which returned year 5700 instead of 5780 — an 80-year error with nothing to signal it. The round trip never caught it because gematriya() only ever emits the regular forms, but תש״ף is how 5780 is actually written in human-authored input.

Fixes #22.

Change

• Map the five sofit letters to their base values (20/40/50/80/90) when parsing — the convention Hebrew dates use, with no ambiguity against existing keys.
• Kept them in a separate sofit2num table rather than adding them to heb2num, because heb2num seeds the num2heb reverse map used by gematriya(); adding sofit keys there would clobber entries like num2heb[20] and corrupt emission. Parsing falls back with heb2num[ch] ?? sofit2num[ch], so number→string output is unaffected.
• HDate.fromGematriyaString inherits the fix, since it delegates to gematriyaStrToNum.

Tests

• gematriya.spec.ts: the six cases from the issue, plus a final/regular equivalence check (תש״ף === תש״פ).
• hdate.spec.ts: end-to-end HDate.fromGematriyaString('א׳ שבט תש״ף') → 1 Sh'vat 5780.
• Updated the JSDoc; the docExamples test auto-verifies the new @example.

Full suite passes, tsc and lint clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_01PdsgtSpAmYU8ChpqUo8fTH

---

Generated by Claude Code