Add a configuration variable for negative label count for the first model

[martinscooper]

Is your feature request related to a problem?

Currently, the system administrator can tweak two system variables to change the amount and type of labels the user has to provide to trigger the training of the models. Those are changed_element_threshold and first_model_positive_threshold.

What is the expected behavior?

Some models need a minimum number of negative labels to perform well, but there is no way to enforce the user to provide them. Add a new configuration variable called first_model_negative_threshold. This new variable will be used to calculate the first model training label requirements.

Additional context (if any):

Both the conditions of the train_if_recommended() function and the get_progress() function.

[martinscooper]

Resolved by #573.