Make optional argmax for y_pred in Confusion Matrix, Precision, Recall, Accuracy

🚀 Feature

Today, the conditions on the input of the Confusion Matrix, (and Precision, Recall, Accuracy in multiclass case) are the following:

    - `y_pred` must contain logits and has the following shape (batch_size, num_categories, ...)
    - `y` should have the following shape (batch_size, ...) and contains ground-truth class indices
        with or without the background class. During the computation, argmax of `y_pred` is taken to determine predicted classes.

Taking argmax on y_pred can be an option if we would like to determine winning class by some other rule. Let's keep argmax as default behaviour if y_pred is (N, C, ...) and do not apply it if y_pred.shape == y.shape and (N, ...).

Start by locating the Confusion Matrix, Precision, Recall, and Accuracy metric implementations and their tests. Check how predictions are handled for (N, C, ...) versus y_pred.shape == y.shape, then verify that default argmax behavior remains unchanged and same-shaped inputs support an alternative winning-class rule.