v0.2.0

[0.2.0] — 2026-07-25

Two additive token groups. No behaviour changes to existing patterns — every 0.1.1
pattern formats and parses byte-identically.

Added

• 12-hour clock tokens h / hh (hour 1–12) and a (AM/PM), valid on
  PlainTime, PlainDateTime and ZonedDateTime. 12:00 AM is midnight, 12:00 PM
  is noon; parsing accepts any case (pm, Pm, PM). This is fixed English text, not
  CLDR data, so the zero-dependency and no-locale guarantees are unchanged.
  • A pattern mixing H with h/a is rejected as contradictory
    (InvalidPatternError).
  • On parse, h requires a and a requires h; an hour outside 1–12 fails loudly.
• UTC offset variants X (±HH, widening to ±HHMM when minutes are non-zero),
  XX (±HHMM) and XXX (±HH:MM). All three render UTC as the literal Z and
  accept Z on parse, normalizing it to +00:00. Existing ZZ is untouched: always
  ±HH:MM, never Z.
  • Sub-minute historical offsets can't be represented in any X form and throw a
    FormatError pointing at ZZ.
• Test suite: dedicated reflect coverage and temporal-sql-style edge-case tests
  (pre-1970 and proleptic/BC years, years ≥ 10000, leap-day validity, DST fall-back
  fold, fractional-second and boundary-time precision), plus 12-hour and offset-variant
  coverage plumbed into the fast-check round-trip properties.
• CI: GitHub Actions workflow — a check matrix on Node 18/20/22/24/26 (26 exercises
  native Temporal) plus a separate attw tarball-gate job.