TO-DO list, from concept to prototype

To be implemented

[x] script: window half size in output filenames
[x] global regression (i.e. non-windowed)
[ ] output in a single netcdf file, with fields
[ ] example data, synthesized or from a real case (e.g. real topography, generate random (smooth) c0, c1 and synthesize correlated observable, masking a real uncorrelated signal)
[ ] unit tests
[x] package
[x] multiprocessing parallelism
[ ] why xarray? the core rolling regression works fine for any array of the same size - we only need to assert for that.
[ ] perform the decorrelation, including a remove-compute-restore "average preserving" strategy (otherwise the decorellation carries an implicit HP-filter side effect)

[ ] aspect ratio (arising from range-azimuth coordinates in InSAR applications)
[ ] managing nans
[ ] output quantities
[ ] predicted part and residuals
[ ] documentation of script behaviour
[ ] module: decorellation without file i/o (pending splitting of script and module)