Reduction for CrossEntropy is by default "none" and not alterable despite documentation

[M-Colley]

While the documentation states that for CrossEntropy, the reduction used is "mean" (https://pytorch-forecasting.readthedocs.io/en/v1.0.0/api/pytorch_forecasting.metrics.point.CrossEntropy.html), the implementation defaults to "none" (see here).

Is this to be expected? For me, this was unintuitive.

Kind regards