Show legend for rasterized overlay

[droumis]

Is your feature request related to a problem? Please describe.

When rasterizing a plot, you lose the ability to show a legend. Rasterize is used when there is a lot of data, so a legend is even more important in this context to visually parse the plots.

Describe the solution you'd like

This FR is to have the code below display a legend.

import pandas as pd
import holoviews as hv
from holoviews.operation.datashader import rasterize
hv.extension('bokeh')

df = pd.DataFrame({'x': [1, 2, 3, 4, 5, 6], 'y': [1, 2, 3, 4, 5, 6], 'source': ['a', 'a', 'b', 'b', 'c', 'c']})
df['source'] = df['source'].astype('category')

curve_dict = {}
for src, data in df.groupby('source'):
    curve_dict[src] = hv.Curve(data[['x', 'y']], 'x', 'y').opts(show_legend=True)

overlay = hv.NdOverlay(curve_dict, kdims='source').opts(show_legend=True)

rasterized_overlay = rasterize(overlay, line_width=5, aggregator=ds.by('source', ds.count())).opts(show_legend=True, responsive=True, min_height=300)

rasterized_overlay

or similar with hvplot:

import pandas as pd
import hvplot.pandas
df = pd.DataFrame({'x': [1, 2, 3, 4, 5, 6], 'y': [1, 2, 3, 4, 5, 6], 'source': ['a', 'a', 'b', 'b', 'c', 'c']})
df['source'] = df['source'].astype('category')
df.hvplot.line(x='x', y='y', by='source', rasterize=True, line_width=5, legend=True, colorbar=False, responsive=True, min_height=300)

Describe alternatives you've considered

alt option: Manual creation of a plot to overlay

• Downside to this approach is you loose the ability to click on an entry in the legend to hide a particular line. But unless each line is rasterized independently, I suppose this is unavoidable (unless clicking an entry was to remake the whole plot).
• There appears to be a util in holoviews categorical_legend from this PR that I can't get to work. It's supposed to create a points element that a user would overlay with the rasterized plot to show the legend. I don't see this documented or used anywhere.
• The census example shows an example of manually creating such an overlay.

alt option: rasterize one curve at a time and then overlay

• It looks like one person on discourse figured out a solution to this by rasterizing individual curves and then overlaying

Additional context

• Relevant discourse thread
• PR awaiting review to alter the categorical_legend util.