simple logging class

[GabrielBG0]

Adds a class that handles logging metrics during training, val and testing.

[otavioon]

Nice work, @GabrielBG0. I made a few modifications:

1. Since batch metrics are typically common to any model, I decided to remove the LoggedSupervisedModel and instead added train/test/validation metrics directly to SimpleSupervisedModel. It's now a dictionary with the metric name as the key and the metric implementation as the value.
2. As previously decided, we're sticking with the torchmetrics.Metric API.
3. I've managed to enable metrics to run on both CPU and GPU.
4. I've added kwargs to all models inherited from SimpleSupervisedModel to allow passing metrics.
5. Added a test of metrics usage using U-Net.

[otavioon]

A few modifications to SimpleSupervisedModel to allow metric logging using torchmetrics.Metric API