OPTIM: Speeding up EOReader

[remi-braun]

It could be really useful to use external libraries to speed up EOReader and ensure the clusterization works properly:

• Optimize the code using libraries such as numba, xarray-geospatial...
• Use WarpedVRT instead of reprojecting data (VHR...)
• Use directly S3 files when possible (wait for cloudpathlib to work properly with fsspec or maybe use universal_pathlib) (-> small optim, most of the time S3 files are used, with the Path converted to string -> but better for the code's reading)
• Hope that one day SNAP will be able to process S3-stored products (right now, it seems still impossible)
• Make Dask really work (even if ENH: Dask support #4 is closed)
  • Ensure reading and writing is accelerated
  • Ensure everything is as lazy as possible (don't call .data !)
  • Ensure that everything has proper chunks
  • Check if reprojection, geocoding, index computation can be daskified and/or optimized:
    • Spyndex ENH: Replace custom spectral indices by spyndex library #47
    • geocode S3 OPTIM: Use pyresample to geocode Sentinel-3 products #55,
    • reprojection: Use 'odc-geo' to handle dask reprojection #115
    • Don't load data into memory before computing spectral indices
    • Use chunks when reading data
• Enforce the window reading (ENH: Enabling the support of windows and AOI when loading bands #25)
• Don't use SNAP for SAR data (ENH: [SAR] Don't use SNAP #2)

[remi-braun]

WarpedVRT optimization has been implemented in 9aaf672