Optimize display of large DynamicMap parameter space

[philippjfr]

As described in #2645, a DynamicMap with a large parameter space can be very slow to initialize, this is because we have started to process its keys more consistently but did not take into account the fact that the DynamicMap parameter spaces can be much larger. This now avoids a number of computations that are not necessary and optimized some others:

def test(t, z, cmap, vmin, vmax):
    return hv.Image(np.random.rand(100, 100))
    
dmap = hv.DynamicMap(test, kdims=['t','z', 'cmap', 'vmin','vmax']).redim.values(t=range(10), z=range(10) ,cmap=['viridis','Greys'], vmin=range(100), vmax=range(100,200))

This DynamicMap defines a parameter space with 2 million keys, previously this would take 52 seconds to display, with this PR this drops to 0.5 seconds.

[philippjfr]

@jlstevens Another good reason to get a 1.10.3 release out.

[jlstevens]

Don't often see a use for itemgetter and I'm trying to think whether you really need it here. I don't immediately have a better suggestion so it is probably ok...

[philippjfr]

itemgetter is about 100x faster than the previous tuple comprehension.

[jlstevens]

That seems to be a good reason. Could you put in a comment to that effect?

[philippjfr]

Sure, that said itemgetter is used frequently throughout utils, so I'm not sure this one deserves a special note.

[jlstevens]

I've made one comment and I'm not sure whether you are done with it, but at a glance this PR seems fine.

[philippjfr]

Ready to merge.

[jlstevens]

Looks good. Merging.