New features

Radiometric terrain correction (or flattening) is now available in the SLC processor
Our algorithm differs from the SNAP implementation. This is actually a modified version of the radiometric normalization algorithm described in SNAP's terrain correction documentation (see https://step.esa.int/main/wp-content/help/versions/9.0.0/snap-toolboxes/org.esa.s1tbx.s1tbx.op.sar.processing.ui/operators/RangeDopplerGeocodingOp.html). Two things are different:
- Instead of the sine of the projected incidence angle, the tangent is computed to comply with the gamma nought convention (as in Small's algorithm).
- The simulated backscatter is regridded and accumulated in the SAR geometry to account for the many-to-one and one-to-many relationships.
Shadows are detected and set to NaNs in the normalized image.
Added tutorial notebooks to demonstrate feature usage.

Bugfixes and improvements

Add libgdal jpeg dependency to work with the new version of rasterio.
Temporary fix: latest pyproj version was introducing a bug in the computation of the DEM XYZ coordinates. Force to use the previous version.
Refactored both InSAR and SLC processors: we do not use child processes anymore and set the GDAL_CACHEMAX value to a small number. Benchmarks showed smaller memory usage than previously and no performance loss.
Fixed some exceptions in the processors (the raise keyword was missing leading the exception to not be triggered).