classification_metrics recall & precision not for binary case

[stewu5]

Current recall and precision calculation in the classification_metrics function is only for multi-class cases, because the average option is set to be "weighted", which becomes a weighted recall or precision for even the binary classification case. So, in the binary case, it is not the true recall / precision.

[TsumiNa]

240 See here. I'd like to add a option to control how to calculate scores. This is an example shows how to use new function.

trainer = Trainer(
...
 ).extend(
    Validator(
        lambda y_true, y_pred: classification_metrics(y_true, y_pred, average=None),
        early_stopping=30, trace_order=5, warming_up=warming_up, accuracy=1.0, micro_f1=1.0, micro_precision=1.0, ...),
)