Add optimized version of distance computation

[msorvoja]

Adds an optimized version of distance computation using spatial indexing (scipy.spatial.cKDTree) for nearest neighbor search and Numba (njit decorator) for more efficient computation of distances. Old and new implementation are compared in notebooks/testing_distance_computation.ipynb.

[nmaarnio]

The final optimized version does not use cKDTree, only Numba.

Performance comparison of old and new version:

Geometries: Line dataset with 85 feature

Old version

• Raster with 2.2M pixels: 42s
• Raster with 13.9M pixels: 367s (~6min)
• Raster with 55.8M pixles: ?

New version

• Raster with 2.2M pixels: 4s (10x speedup)
• Raster with 13.9M pixels: 12s (30x speedup)
• Raster with 55.8M pixles: 56s

Geometries: Polygon dataset with 8 features

Old version

• Raster with 2.2M pixels: 22s
• Raster with 13.9M pixels: 160s
• Raster with 55.8M pixles: ?

New version

• Raster with 2.2M pixels: 2s (11x speedup)
• Raster with 13.9M pixels: 3s (53x speedup)
• Raster with 55.8M pixles: 8s

Additional Distance to anomaly test:

• New version with 13.9M pixel raster: 7s
• Old version with 13.9M pxiel raster: did not finish in 620s (~11min) => at least 90x faster for this dataset, in practice makes Distance to anomaly from unusable to usable and fairly quick

[msorvoja]

@nmaarnio Tested with IOCG point, line and polygon data, and the results look good to me!

[nmaarnio]

There was a bug @Mtk112 noticed that occurred for some numpy and rasterio versions on Windows at least. It should be fixed now.

This PR should be ready to be merged now 🚀