Refactor before stable version

[ktagowski]

Issues:

• Shorten library imports #181
• Move dataset downsampling from HuggingFaceDataModule to separate transformation #182
• Configure optional dependencies in pip #183
• Refactor tests #117
• Refactor LightningTask #153
• Improve logging and replace 3rd party progress bars #162
• Remove flair embeddings for transformers #167
• Provide error messages while passing static embeddings to lightning pipelines #169
• Add configurable CUDA Version for package installer #111
• Update gpu Dockerfile #170
• Unify predict_subset along different pipelines
• Refactor HPS pipeline #209
• Add separate issue to mypy dataclasses with abstract methods issue and add #TODO annotations in files instead of comments
• Unify default seed as default variable among all methods
• Improve naming and behaviour of input_column_name in pipelines as currently it can be both str and list
• Resolve multiple loads of dataset in Lightning Datamodule
  Cleanup:
• Setup appropriate defaults for scripts and README: embedding, dataset, parameters etc.
• Correct organization of modules

Ideas:
- [ ] Consider refactoring embedding pooling to transformations

• Consider moving all type aliases to types.py file
• Consider prefixing boolean variables with should_ or if_ etc.

[laugustyniak]

@ktagowski do we want to update it or close and review issues separately?

[ktagowski]

@laugustyniak Let's move to a new one with a review of the issues, and close that one

• Some issues are important from a user perspective, therefore, should be prioritized
• Code improvements can be handled by one refactor.
• Other issues should be closed or moved to future work