removing evil test_set query and updating validation to output metrics more often

full training run completed, created combined dataset tsv, validation code must be updated still

first full fine-tuning run, resulting model stored in finetuned-model-16

Ran first 1 epoch training run, trying to extract private data, no luck yet :(

bugfix in finetuning script, added gitignore

adding tennis sqlite data, basic utils and training/testing python notebooks