Check pip requirements/dependencies

[howff]

Had to use pip install --prefer-binary because Guneet's Windows PC had some odd configuration which meant that pip was pulling in packages marked as pre-release, which were source-code only (no compiled binary package available yet) and a lack of a Windows compiler meant the packages could not be installed.

Cannot use --only-binary=:all: because things like pytesseract are only source (which is fine, doesn't need to be compiled).

However Daniyal has reported dependency problems with things like matplotlib, see https://git.ecdf.ed.ac.uk/SMI/service/-/issues/137#note_157855

[howff]

Multiple problems are now cropping up, as the various dependencies gradually change.

e.g.

• torchmem.py - crashes when run on CPU with the GPU pytorch package installed (the Exception has changed)
• newer pytorch causes Illegal Instruction (core dumped) in easyocr (try different pre-compiled models, or downgrade pytorch)
• pytorch CPU-specific packages should be installed if no GPU, not sure how to write that in requirements.txt
• numpy v2.0 - causes problems with the bitmask and signed integers
• pydicom 3 is now out - deid only compatible with pydicom 2
• pydicom jpeg libraries - may have system component dependencies
• numpy / scipy / scikit - pip sometimes fails to resolve dependencies, either pulling in a source package which won't compile, or pip gets into a loop and crashes after 200,000 iterations
• fastDamerauLevenshtein only available as source, it's an optional dependency but I don't know how to write optional in requirements.txt

Sometimes it works if you install packages one at a time, but fails if you pip install -r requirements.txt