ENH: Add segmentation metrics

[NickleDave]

I started this here: https://github.com/vocalpy/vak/tree/add-segmentation-metrics using the segmentation metrics built into VocalPy

But if we're going to use torchmetrics, really we should

• [ ] re-implement all the functions using pure torch
• [ ] do this in a functional module
• [ ] then have these classes

This is duplication but I think in this case worth it to be able to have models work purely with tensors instead of flipping back and forth b/t torch.tensor and numpy.array