Training Pipeline working on the FinQA dataset

[iusztinpaul]

We implemented the training pipeline on the FinQA dataset as a temporary solution.

To try it out, install the virtual environment, fill in the credentials, and run it:

make install
# Fill credentials
make train_local # or make train_beam
make infer_local # or make infer_beam

NOTE: Soon, we will adapt it to our own semi-automated Q&A dataset.