Speed up prediction and outlier detection.

See merge request !9

Using threading for parallel processing speeds up prediction by a factor of 100: most of the time consumed was on passing data to the processes. Switched from the EllipticEnvelope to an IQR-based outlier detection method -- faster and does not assume Gaussianity. Since this happens after the PCA, the data is already decorrelated.

Added Wiener deconvolution.

See merge request !8

Improve fit

See merge request !7

Use per channel PCA for consistency check. Original fit model has better uncertainties, when compared to ARD regression: switching to it.