Regression bug fix of scatter2d traces not to change other traces colors

[archmoj]

Fixes #3232 and https://github.com/plotly/phoenix-integration/issues/232 by disabling unstable palette color algorithm in regl-scatter2d.
Although some improvements are made in another branch and PR; that still requires more rewrites and possibly passing two palettes to the shaders.
While this could be a very annoying bug for Plotly users, this PR disabled the buggy palette mechanism in the module so that it may be fixed in the next minor version.
Also few other minor improvements are made in the regl-scatter2d module.
@etpinard
@alexcjohnson

[archmoj]

gl-vis/regl-scatter2d#16

[etpinard]

Have you ran any benchmarks comparing with-palette and w/o palette performance?

I'd be ok (not sure I'm @dy would be though) merging this thing if performance slows down by "a little", but I'd like know by how much.

[archmoj]

old: 4730.064ms
new: 4733.689ms
code:

var i, RGBA = [], X = [], Y = [];

for (i = 0; i < 1234567; ++i) {

X[i] = Math.random();

Y[i] = Math.random();

RGBA[i] = [Math.random(), Math.random(), Math.random(), Math.random()];

}

traceType = "scattergl";

console.time(traceType)

Plotly.newPlot(gd, {

data: [{

"type": traceType, "x": X, "y": Y,

"mode": "markers", "marker": {"size": 1.5, "color": RGBA}

}],

layout: {width: 1200, height: 800}

});

console.timeEnd(traceType);

[archmoj]

And using a code that only generates red values (i.e. 0 - 255 range) I've got:
old: 3668.412ms
new: 3689.617ms
code:

var i, RGBA = [], X = [], Y = [];

for (i = 0; i < 1234567; ++i) {

X[i] = Math.random();

Y[i] = Math.random();

RGBA[i] = [Math.random(), 0, 0, 1];

}

traceType = "scattergl";

console.time(traceType)

Plotly.newPlot(gd, {

data: [{

"type": traceType, "x": X, "y": Y,

"mode": "markers", "marker": {"size": 5, "color": RGBA}

}],

layout: {width: 1200, height: 800}

});

console.timeEnd(traceType);

[archmoj]

Codepen demo 1 & demo 2 using this branch.

[dy]

The benefit here is not performance, but memory optimization - for <= 256 colors we require a single uint8 per-color, and for buffer that is uint32. We had that issue long ago with gl-scatter2d, when faced memory limitation.

[etpinard]

Thanks for note @dy !

@archmoj could you take a screenshot of the task manager (open it with shift+escape)

on master and on your branch to compare memory consumption?

[archmoj]

[archmoj]

Closed in regard with #3328. But I prefer not to delete the branch for further improvements.