Remove python dependencies

[hollandjg]

When working on the pipeline, I was having trouble with building the docker because of the PyCall dependency. I wanted to see how easy it was to replace PyCall with PythonCall which has a better way of handling dependencies.

Turns out it's also hard:

• PythonCall doesn't have an obvious way to import a custom file (latlon.py)
• We have to package that separately, and the import it using Pip in Conda.
• Using Pip in Conda reinstalls the dependencies from the package installed using Pip.
• We'd need to make latlon.py into a conda package and push it to a registry to make this easier.

I don't think we should merge this PR – it just makes things complicated in a different way. I think we should do our best to remove the python dependencies from this package.

• LatLon: Test functionality from JuliaGeo libraries Rasters.jl and Project.jl to replace Python code in getlatlon.py #457
  • We could replace it with something based on Proj.jl (preferred)
  • We could have a separate command-line tool which does the reprojection and read the values in from a flat file
• Scikit-image: region props:
  • We could have a small command line tool based on a part of ebseg which takes an integer labeled image and produces the region props as we need them
  • We could have a small command line tool (new) which takes in a binary mask and labels the contiguous blocks using integers, and then feed that into the first command line tool.
  • Then we read all the region props back into the julia pipeline
• Scikit-image: Exposure correction:
  • we could have a small command line tool (new) which does the exposure correction on a full image, and then pass that image back into the main pipeline.

I tried to get around this by updating the Conda environments, but found that by default the PyCall installs python3.12 which isn't compatible with our dependencies.

• PyCall now installs Python3.12 by default which is incompatible with some of our python dependencies #465

[danielmwatkins]

We might be able to take advantage of the relatively new Julia wrappers for OpenCV: https://github.com/JuliaImages/OpenCV.jl

This would potentially remove the need to call Python for the skimage stuff, however there may be new issues if the OpenCV bindings don't work well on all systems.