Add eval of pretrained transformer models using haystack. rename embedding folder

[tholor]

• add evaluation script for plain vanilla bert-base-uncased, sentence-bert and a multilingual sentence-bert
• benchmarking a few pooling methods + extraction layers
• all results in https://public-mlflow.deepset.ai/#/experiments/55