Dataset refactoring

Summary

As a user, I want to rewrite the dataset in xarray/rioxarray, so we can be more flexible in train and validation.

Description

List of steps:

1. clipping via rasterio of rasters from stack to select sample points in the vineyards
2. use the stack with (ndvi,meteo,soil,geo) at 4 different resolutions, using multiple netcdf coordinates and sampling with "sel" using "nearest" interpolation

Notes:

• ndvi and meteo have dates as channels, but there's no need to interpolate, the whole array is fed to the training loop
• geo has 6 channels: one for each geo property
• soil has 1 channel