Add precipitation histogram to prognostic run report

[oliverwm1]

Useful to show a histogram of precipitation on prognostic run reports. Example report HERE.

Significant internal changes:

• Added function to compute histograms of arbitrary variables. For now, just doing surface precipitation rate.
• Create new process_diagnostics.html page to the prog run report. Moved diurnal cycle there, and added plot of precipitation histogram.
• Added metric to compute percentiles from above histograms. For now, computing 25th, 50th, 75th, 90, 99, and 99.9th percentiles. e.g.:

"percentile_25/total_precip_to_surface": {
      "value": 0.3747434426709828,
      "units": "mm/day"
}

Sped up report test by only creating report for a single run (since no longer require >1 run for report generation).

• Tests added

[oliverwm1]

We switched to using the vcm.dump_nc version of this a while ago, so this func is unused.

[oliverwm1]

I found using np.histogram much faster (and simpler code-wise) than doing an xarray groupby.

[oliverwm1]

deleted unused argument run_attr_name

[nbren12]

Looks great. Apart from some minor comments below, I had some thoughts

1. I was pretty confused about why the metric functions returned xr.Dataset rather than floats, before realizing that the to_dict function looks at the units and values. It would be clearer if the metrics functions returned dataclasses with .value and .units attrs, so that we are passing the minimal amount of data around.
2. It would be great to enhance the visibility of the scalar metrics, so that they are amongst the first things we see. e.g. by showing a table instead of the histograms. The histograms are kind of useless and ugly IMO, especially for a single run report.

[oliverwm1]

I was pretty confused about why the metric functions returned xr.Dataset rather than floats, before realizing that the to_dict function looks at the units and values. It would be clearer if the metrics functions returned dataclasses with .value and .units attrs, so that we are passing the minimal amount of data around.

yeah I agree the xr.Dataset is a bit overkill. We could improve this in another PR.

It would be great to enhance the visibility of the scalar metrics, so that they are amongst the first things we see. e.g. by showing a table instead of the histograms. The histograms are kind of useless and ugly IMO, especially for a single run report.

ouch, poor histogram ;) It would be helpful to have the verification data on there, I agree. The example is particularly noisy since it's a short run. But yeah, a table showing the percentiles for each run would be quite useful.

[nbren12]

ouch, poor histogram ;)

sorry. I didn't mean your histograms, i meant the bar charts with the metrics on the main page. The histograms are great!

[oliverwm1]

ouch, poor histogram ;)

sorry. I didn't mean your histograms, i meant the bar charts with the metrics on the main page. The histograms are great!

haha sounds good. I agree the bar charts are hard to parse. I'll work on adding a metrics table with the things we actually care about.