Mlflow+EarlyStopping

[Anya497]

• Create EarlyStopping and add it to the training pipeline.
• Separate two types of learning with study's URI in config file. If you want to train model with concrete hyperparameters you need to specify URI of optuna's study that contains information about the best trial with hyperparameters you interested in. Also you need to add to config model's weights URI which correspond to chosen model's parameters. Otherwise optuna's hyperparameters tuning will be started.
• Add validation type to config. Move number of servers to configuration of validation with SVMs.
• Speed up loss validation with batches.
• Add logging with MLFlow.