GlucoseValue.decayEffect currently accumulates the decay step-by-step starting from startDate.dateFlooredToTimeInterval(delta). That makes the effect value at any future absolute timestamp depend on which delta-sized simulation bucket the sample's startDate happens to fall into — so two CGM samples a fraction of a minute apart can produce noticeably different effect curves at the same future point, even though the math is supposed to be smooth.

This PR reformulates decayEffect as a closed-form quadratic f(t) = a*t² + b*t + c with t measured from the actual sample timestamp. For samples aligned to delta boundaries the two formulations are mathematically equivalent; for unaligned samples the new formulation removes the bucket-boundary discontinuity.

This is a port of LoopKit/LoopKit#556 by Moti Nisenson-Ken, brought over to the LoopAlgorithm package where decayEffect now lives.

Changes

• Sources/LoopAlgorithm/LoopMath.swift: reformulated decay computation.
• Tests/LoopAlgorithmTests/LoopMathTests.swift: new testDecayEffectIsContinuousAcrossSimulationBoundary exercising the continuity property.
• A handful of existing fixture-calibrated tests had expected values that were tied to the old accumulation; those were re-pinned to the new outputs. Per-prediction drift is on the order of 0.1 mg/dL, which is in line with the original PR author's note that downstream fixtures would need follow-up.