Pass customdata through to scatter calcdata

[rreusser]

This PR adds a customdata attribute to scatter traces that is passed through and assigned to the data property of the corresponding calcdata entry.

To do: verify the manner in which this is (or is not?) actually passed all the way through to the dom element.

ping @alexcjohnson @etpinard

[alexcjohnson]

Looks reasonable to me. Do you also want to:

• add a class like plotly-customdata to the elements that actually have d.data (I think scatter/style.js is the right place (do we need to call out text or just path.point?) so the API doesn't need to know anything about our selectors to find the right elements to look for d.data on.
• explicitly test passing through an object, not just a string, as that's probably what we'll need in practice.

[rreusser]

I've added a plotly-customdata class that's set if data is anything except undefined or null. I could probably avoid the loop if customdata wasn't even provided. It didn't feel like a nice fit in the scatter tests, but I didn't really want to unit-test an intermediate API for this since it certainly doesn't need to be implemented in the style pass.

[alexcjohnson]

Nice. Yes, I think it'd be worthwhile to avoid the loop in the (overwhelmingly common) case of no customdata.

[alexcjohnson]

Beautiful test. I agree, end-to-end is the right way to test this particular beast.

[alexcjohnson]

Just a little style point: pt makes it sounds like it refers to a single point, but it's all points so I'd change it to pts. And elsewhere cd refers to a whole calcdata array so this tripped me up for a sec - maybe change it to pointData or something.

[rreusser]

Caveat I just can't ignore: animation. Going to add that to the test to ensure these are properly updated. Since transitions don't work through the style pass, an animation update won't update these. Fixing. This might be reason enough to move this to the plot step instead.

[rreusser]

Okay, I've updated the PR (in a test-driven fail-first manner that actually tests what it claims to 👍). It unsets customdata via an animation that doesn't replot in order to verify that it's actually set/unset/updated correctly. This required moving it to plot.js which is pretty straightforward and doesn't really affect anything, to the best of my knowledge. I'm currently content with this.

[alexcjohnson]

Very nice. 💃