[BUG]: Color Translation is Costly

Description

When rendering a scatter trace, we run rgb(color_string) and parse(color_string) four times for every mark. For the vast majority of scatter charts, this is wasteful re-computation.

Surprisingly, this string parsing appears to be a serious bottleneck. Removing it seems to provide a 2x speed improvement for charts with large numbers of marks.

Screenshots/Video

Steps to reproduce

Run npm start.
In the the console run:

const gd = document.getElementById('graph');
const n = 1e5, x = new Float64Array(n), y = new Float64Array(n);
for (let i = 0; i < n; i++) { x[i] = i; y[i] = Math.sin(i / 500); }
const runs = [];
for (let k = 0; k < 5; k++) {
    await Plotly.purge(gd);
    const t = performance.now();
    await Plotly.newPlot(gd, [{type: 'scatter', mode: 'markers', x, y}],
        {width: 900, height: 600}, {displayModeBar: false});
    runs.push(+(performance.now() - t).toFixed(1));
}
runs.sort((a, b) => a - b);
console.log('median', runs[2], runs);

See the numbers. Then test with the memoization PR I've put up and notice the lower numbers.

Start with the scatter trace rendering path that calls rgb(color_string) and parse(color_string); compare it with the memoization PR mentioned in the report. Reproduce the 100,000-mark benchmark from the issue, then confirm the optimized run retains rendering behavior and improves median timing.