colour: the ±12 decoy controls read dead zeros and inflate the excess statistic

Found during PR #162 review (pre-existing, orthogonal to the width change): the ±12 deg colour decoys read a hard 0.00% on every lag because reach defaults to 8 and they lie entirely outside the profile window — and excess = seam − mean(all four decoys) averages the two dead zeros in. On the review's test view this inflates the 32px excess from 4.47% to 5.61%.

Fix directions: clamp decoys inside the profile window, or exclude out-of-window decoys from the mean, or widen reach with the decoys. Filed under the owner's standing issue authorization.

No files are named, so first find the colour profile/lag metric code path that computes seam and excess from the four decoys. Reproduce the reported behavior from the PR #162 review scenario and identify where the ±12° decoy values become hard 0.00%. Adjust the decoy handling so out-of-window items are clamped, filtered, or reach is widened, then confirm the 32px excess statistic is no longer inflated.

Escrita pelo modelo de indexação a partir do texto da issue.