Add support for Numpy/Scipy functions

It is possible to compute additional "spectral" feature per date using the feature Sensor.additionalFeatures. However, it is not possible to add "temporal" features, i.e., feature computed on the temporal stack for one/each spectral band.

To let more flexibility, iota2 should offer the possibility to the user to define processing involving any subset of the spectral and/or temporal features. One possibility is to allow the user to provide a scipy/numpy function that will be applied pixelwise during the data generation (for learning, classification etc...).

This is something similar to #87 (closed) prediction step. But the computation will be done before learning the model, to add new feature to the data cube.