Support for conformal prediction

[Geethen]

Integrating conformal prediction based on mapie library to create prediction sets.

Current approach:
The data (train, val and cal data) and inference image is converted to local objects.
A portion of of the initial dataset is set aside as the calibration data (used to calibrate a conformal predictor). A random forest model is trained on both the training and validation data. Thereafter, the calibrated predictor is used to predict the prediction sets on the inference image. The inference Image is downloaded locally using the geedim package, thereafter, the image is processed in chunks using rasterio.