metric grid in the predict_model function of time_series

When the predict_model function is used without the fh parameter the metric grid makes sense as it is evaluated on the test set. However, when fh is not None metrics doesn't make sense since it is expected that predict_model used with fh None means the user is generating predictions in the unknown future. The test set is an exception that's why predict_model(model) just by itself is valid only when fh=None.

Please change this behavior as such when fh is not None the metric evaluation shouldn't be performed or shown (leave it up to you).

Example: