Images_to_samples.py (hdf5s) --> data_to_tiles.py: add gdal library in conda

[remtav]

Onboarding solaris requires GDL to use gdal as a python library (not just from the OS). This will require us to update environment.yml to create a functional conda env with this new dependency.
For context, see #215

[ymoisan]

I gather we'll keep using rasterio right ? If so, we should investigate how to better use rasterio's gdal bindings and avoid potential issues.

[remtav]

I gather we'll keep using rasterio right ? If so, we should investigate how to better use rasterio's gdal bindings and avoid potential issues.

Thank for raising this flag. Digging deeper into the solaris utilities that will be necessary for us, we can easily remove gdal as a python dependency. Solaris basically implemented the use of gdal datasets side by side with rasterio dataset. 2 examples (only 2 scripts use from osgeo import gdal):
https://github.com/remtav/geo-deep-learning/blob/solaris_tiling/solaris_gdl/raster/image.py#L31
https://github.com/remtav/geo-deep-learning/blob/solaris_tiling/solaris_gdl/utils/geo.py#L151

If we decide to remove all implementations of gdal.dataset, etc., we can remove gdal as a dependency and close this issue.

[remtav]

Solaris v0.5 will drop the gdal dependency.