model monitoring

[isabelizimm]

adds

• plot_metrics + tests
• compute_metrics + tests

With pinning in separate PR to round out all the monitoring! 🎉

[machow]

I can take a closer look in a little bit, but one quick thought -- can we add some extra tests of the rolling window? Vetiver R has the luxury of leaning on the slider package 😭!

Looking at the code, it might be good to test against these potential bad things:

• the rolling window accidentally produces the same dataframe for each window
• the metric doesn't get applied per window, so returns the same result for each window

Another quick thought. a lot of R functions can use the actual variable name in the output. vetiver-python matches this using .__qualname__. But there are often cases where .__qualname__ doesn't exist, like object instances and relatedly functools.partial. I wonder if it'd be useful to provided a output naming escape hatch, like allowing a dictionary of metrics metric functions.