Add support for HoloViz/HoloViews plots

[nstrayer]

Addresses #3880

What

This PR adds a new service that intercepts messages from holoviz plots so they can be replayed when a plot is rendered into the plots pane.

Why

This is necessary because holoviz sends preliminary messages that load assets for displaying plots and these messages need to be sent to the plot webview along with the plotting message when rendering the plot or else various assets will not be available.

This is not a problem in notebooks because all output is rendered into a single large "backlayer" webview and thus the preliminary messages are properly "remembered."

This is specific to holoviz/holoviews plots right now but there is a decent chance this pattern of remembering prerequisite messages and replaying them to the output webview will be useful for other libraries/situations.

Limitations

Right now we store the prereq messages in memory on the UI thread. This is nice and lightweight but has the problem of forgetting those messages if the user reloads the front end. Since hvplots doesn't replay the asset loading messages on reload the plots won't work until the session is restarted. We could do something like make the storage of prereq messages use a non-ui layer cache, but I'm not convinced it will be a big issue.

Results

hvplots now work in the plots pane

Behavior in notebooks should be unchanged

QA Notes

This code will create a simple hvplot scatter plot. It should work in the plots pane.

import polars as pl
import hvplot
import hvplot.polars
pl.DataFrame(dict(x=[1,2,3], y=[4,5,6])).hvplot.scatter(x="x", y="y")

The same code should work in a notebook. (Although even before this PR I noticed sometimes it was flakey which I think is due to how hvplots loads assets).

[seeM]

Very nice @nstrayer! I went through a bunch of plots from https://hvplot.holoviz.org/ and they seem to work well in the console and notebooks! They also reload correctly when changing the active pane or collapsing the plot pane. Also confirmed that pre-existing plot types continue to work (Python/R Plotly, Python ipywidgets).

I noticed two issues which we can address separately:

Plot widths

I noticed the plot is not sized to fit the plots pane. I'm not actually sure what determines the size. It seems like it's being hardcoded to a width that assumes its in a notebook interface? Since the width seems to be the same for different plot types, in the plots pane and the notebook, and also in their docs. Maybe we can make a note in #4245.

Plotly backend

IIUC hvplot can use different plotting "extensions" with Bokeh being the default and working well in this branch. However, Plotly does not seem to be working yet, we can track that separately if needed.

import hvplot.pandas
from bokeh.sampledata.penguins import data as df

hvplot.extension('plotly')

df.hvplot.scatter(x='bill_length_mm', y='bill_depth_mm', by='species')

[seeM]

Not sure if its necessary, but should we clear the map too?

[nstrayer]

Probably!

[nstrayer]

Actually, I guess this will automatically get garbage collected. Assuming dispose() only gets called when the service is truly being disposed of.

[seeM]

I'm not actually too sure what this affects.

[timtmok]

What's the reason for using a NotebookOutputPlotClient?

[seeM]

Code changes look good, but I seem to be getting an empty plot on the latest commit when I run:

import polars as pl
import hvplot
import hvplot.polars
pl.DataFrame(dict(x=[1,2,3], y=[4,5,6])).hvplot.scatter(x="x", y="y")

[seeM]

Looks good, and working again!