🚀 Feature

I suggest to add new options regarding average parameter of Precision and Recall metrics.
Currently they have two options which are average==False and average==True. I propose to change average options to the following:

• False: return per class/label metric for multiclass/label input and simple precision for binary input (default option). This is equivalent with average=None of Scikit-learn for multiclass/label input and average=binary for binary input.
• micro: for multilabel, computes metric using all samples of all labels altogether(equivalent with micro of Scikit-learn). Incompatible with binary and multiclass inputs, as it would be equivalent with Accuracy and user can use that metric.
• weighted: It computes metric for every class/label then returns weighted average of them using number of samples in each class/label. Recall does not have this option because weighted recall is equivalent with accuracy. Equivalent with weighted of Scikit-Learn.
• samples: just for multilabel input. It computes metric for every sample then returns their average. Equivalent with samples of Scikit-Learn.

I did not put macro option because user can instantiate metric with average=False then call mean on it to become macro metric.