v0.2.2

Fixed
RANSAC sampled the wrong events. The random sample was drawn against the quadrant-subset size but used to index the full dataset, so initial models were fit on arbitrary rows instead of the upper-right-quadrant events the algorithm targets. Sampling now uses the actual quadrant row indices.
Quadrant cutoff was miscomputed. The 50% threshold used half the data range ((max − min) / 2), which is only correct when the minimum is 0; wrong for the offset/negative values typical of asinh/biexponential-transformed data. It’s now the midpoint of the 1st–99th-percentile range, which is also robust to a few very bright events that previously skewed the cutoff.
Inlier sets are now combined with union() so the seed sample and residual-based inliers are consistently indexed.

Added
Input validation on formula, data, and numeric parameters.
Graceful skip (returns NULL with a warning) when too few events fall in the quadrant, instead of erroring inside sample().

Impact: aggregate calls from 0.2.1 and earlier should be re-run; results will differ (and improve), especially on transformed data and channels with bright outliers.