No scattergl clustering below 1e5 pts

[etpinard]

An attempts at fixing #2334 by skipping point clustering (done by point-cluster) for scattergl traces with less than 1e5 pts.

Previously, we were subject to the regl-scatter2d default value of 1e4 meaning scattergl traces with more than 1e4 pts but less than 1e5 pts (our TOO_MANY_POINTS constant) were clustered from regl-scatter2d whereas traces with more than 1e5 pts were clustered from Scattergl.calc.

With the PR, I noticed that panning is slightly slower for graphs 1e4 < pts < 1e5, but otherwise performance appears the same. I think this small performance lost is worth it to get the correct rendering and to make us call point-cluster from one place i.e. from Scattergl.calc.

Here are some updated codepens from #2334:

• https://codepen.io/etpinard/pen/RdwrJy
• https://codepen.io/etpinard/pen/Ygzwve
• https://codepen.io/etpinard/pen/VRwedy?editors=1010

We may eventually still want to lower the maxDepth clustering option as @dy has suggested in #2334 (comment) to potentially correct rendering of traces with > 1e5 pts, but I'd like to wait to find a currently failing example before doing so. We could also try to bring back clustering for 1e4 < pts < 1e5 traces together with a lower maxDepth down the road.

cc @plotly/plotly_js I'd like to slide to in the 1.45.0 - as this PR may have consequences I don't want to release in a patch version.

[archmoj]

Awesome fix!
Before
After
Thanks @etpinard.
Concerning too many line numbers in gl2d_no-clustering.json and also in order to improve the mock and baseline test with points with opacity, would you please replace it with the data printed to console here?

[etpinard]

@archmoj I like the way gl2d_no-clustering.json easily demonstrated the fix (i.e. one pt was missing, now it is not), so instead of replacing it, I simply added a new mock in f6e8918

I hope you don't mind.

[archmoj]

Excellent! Thanks.
💃