fix: test skipped-model membership with a Set, not in on an array

[spokodev]

Problem

The constructor nulls a model argument that is one of the deliberately-skipped models (keyword, gray, hex) so it falls back to rgb — these have no array representation. But the guard never fires, so passing them as the explicit model produces garbage or a crash:

Color([128, 128, 128], 'hex').string();       // "#80NANNAN"   (expected "rgb(128, 128, 128)")
Color([128, 128, 128], 'hex').rgb().string();  // "rgb(17, 34, 136)"
Color([128, 128, 128], 'keyword').rgb();       // TypeError: cssKeywords[keyword] is not iterable
Color([128, 128, 128], 'gray').rgb().string(); // "rgb(255, 255, 255)"

Cause

const skippedModels = ['keyword', 'gray', 'hex'];
...
if (model && model in skippedModels) { // `in` on an array tests indices '0','1','2', never the names
  model = null;
}

The in operator checks an object's keys; on an array those are the numeric indices, so the guard is dead code. The model-method loop lower in the same file already uses the correct test, skippedModels.includes(model) — confirming the intent.

Fix

Make skippedModels a Set and use .has() at both sites. (A plain .includes() at the constructor would trip the repo's own unicorn/prefer-set-has xo rule, so a Set is the clean form for both.)

Verification

• New test in test/index.js; the skipped models now fall back to rgb — fails before, passes after.
• npm test (xo + tsd + mocha): 31 passing, lint clean.
• Fuzz: across all valid models × 700,000 random inputs, output is byte-identical to the current release (the change is behavior-neutral for every real model); the three skipped models now match the no-model rgb fallback (60,000 checks, 0 still broken). Reachable only via the explicit model argument with array/number input, so normal string/object usage is unaffected.