Reduce transitive dependencies from sentence-transformers

[joshdevins]

When we install sentence-transformers in our pytorch target, we pull in all of the transitive dependencies of that library which include many libraries that we will never need. Excluding these transitive dependencies will reduce eland's overall installation footprint and consequently the Docker image size. This simplification can also help reduce dependency conflicts such as #408.

'tqdm',
'torchvision',
'numpy',
'scikit-learn',
'scipy',
'nltk',