v1.3.0

Minor Changes

• #85 ff20d49 Thanks @tenphi! - Add contrastLevel — a manual contrast level that turns the two-tier
  high-contrast model into a 0–100 slider. Set it globally via
  glaze.configure(), or per theme / per token / through extend():

  glaze.configure({ contrastLevel: 60 });
  const theme = glaze(280, 80, { contrastLevel: 60 });
  glaze.color('#26fcb2', { contrastLevel: 60 });

  Level 0 reproduces the normal light / dark output and level 100 the
  lightContrast / darkContrast output, bit for bit. Levels in between are
  resolved at that level, not interpolated after the fact: Glaze interpolates the
  three things that make high contrast differ — authored [normal, highContrast]
  pairs, the tone-window widening (light [10,100] → [0,100], dark
  [15,95] → [0,100]), and the AA → AAA / APCA +15 Lc escalation — and feeds
  those through the ordinary resolve. A contrast floor is therefore genuinely
  solved at every level (contrast: 'AA' at level 50 solves for 5.75), and the
  floor is met at every level.

  While a level is set there is no separate high-contrast tier: lightContrast /
  darkContrast mirror their normal counterparts, and a global level turns
  high-contrast output off outright so no exporter emits the tier.
  modes.highContrast goes inert rather than fighting it — it reads as "emit a
  separate high-contrast set when contrast is automatic" — so a build config that
  leaves highContrast: true set keeps working, silently, when a user switches
  their preference from auto to manual. css() keeps its four-string shape with
  the high-contrast strings repeating the normal declarations, so existing
  @media (prefers-contrast: more) wiring keeps working untouched. A level set on
  one theme of a palette leaves its siblings' high-contrast tier alone.

  A color never swaps sides of its base mid-slider. autoFlip's tie-break —
  when both directions meet the floor, take the one nearer the authored tone —
  depends on the target, so along a ramp it would let a color leap across its
  base. The side is now decided once from the nearer endpoint and preferred
  throughout that half of the ramp: a color whose two ends agree never changes
  side, and one whose ends genuinely disagree changes exactly once, at level 50.
  Flipping is only re-ordered, never disabled, so a side that cannot physically
  reach the target still falls back. This is exposed as a new preferInitial
  option on findToneForContrast.

  Un-interpolable tone pairs ([50, 'max'], [50, '+20'], ['max', 'min'])
  switch at level 50 rather than blending across kinds.

  Also exports resolveContrastForLevel(spec, level, polarity?).

  A contrast pair may no longer switch metric. [4.5, { apca: 75 }] now
  throws a validation error: a WCAG ratio and an APCA Lc are different scales, so
  no target exists between them and the two variants are incomparable even without
  a manual level. Pair values inside one metric instead — { wcag: [4.5, 7] } or
  { apca: [60, 90] }. Previously such a pair resolved silently.

  Nothing else changes by default: contrastLevel defaults to 'auto', which is
  today's behavior exactly, and no existing type, signature, or output shape moved.
  As a bonus, a manual resolve runs two passes instead of four.