Function to perform timescale decomposition of any 3D array

This merge requests adds a timescale decomposition function that separates any given 3D array of time and spatial dimensions into its trend, seasonal, decadal and residual components using LOESS smoothing and Butterworth filtering. It can return either variance percentages for each component, or the array's raw component values.

The merge request includes the function itself, as well as some unit tests for the function.