fix: do not ovelap hoverlabels for different tracetypes

[mbant]

This PR attempts a fix for #6917 which reported excessive hoverlabel removal and overlap for plots with both scatter and bar traces.

Behaviour before the fix:

or even

Behaviour after the fix:

JSON file used for the example

{
  "data": [
    {
      "name": "scatter_1",
      "type": "scatter",
      "y": [0, 1, 2],
      "x": ["CAT 1", "CAT 2", "CAT 3"]
    },
    {
      "name": "bar_1",
      "type": "bar",
      "y": [0, 1, 2],
      "x": ["CAT 1", "CAT 2", "CAT 3"]
    },
    {
      "name": "scatter_2",
      "type": "scatter",
      "x": ["CAT 1", "CAT 2", "CAT 3"],
      "y": [0, 1, 2]
    },
    {
      "name": "bar_2",
      "type": "bar",
      "y": [0, 1, 2],
      "x": ["CAT 1", "CAT 2", "CAT 3"]
    },
    {
      "name": "scatter_3",
      "type": "scatter",
      "x": ["CAT 1", "CAT 2", "CAT 3"],
      "y": [0, 1, 2]
    },
    {
      "name": "bar_3",
      "type": "bar",
      "y": [0, 1, 2],
      "x": ["CAT 1", "CAT 2", "CAT 3"]
    }
  ],
  "layout": {
    "width": 600,
    "height": 400,
    "hovermode": "x"
  }
}

[mbant]

#6917 (comment)

@archmoj / @alexcjohnson any reason why this can't be considered yet?

[archmoj]

Thanks very much for the PR.
It is looking good on my side.

[alexcjohnson]

I don't understand this actually - I only see us using crossAxKey as a variable name, not a key in these point objects.

Given that this change fixes the behavior, you've clearly identified that this condition is the one causing this bug, but AFAICT the change is equivalent to just if(topOverlap > 0.01) - and maybe that's right? Maybe the pmin/pmax conditions are no longer useful (which presumably means the comment Only group points that lie on the same axes is also no longer what we want) but that's the question to answer before we move forward with this.

@archmoj I've forgotten the details of this, and anyway it changed recently with the hoversubplots work, but if you can convince yourself that removing this condition entirely is correct in all the cases we care about (in addition to the case being tested here, I'd focus on overlaid subplots with points at the same position on screen but either different x axis, different y axis, or both) then go for it!

[mbant]

100% agreed on the condition. It's been a while so not sure why that was there ...

I pushed a commit to simplify the condition (and thus removing the comment) if that helps with testing / convincing; I will have time to pick this up again in the following days so @archmoj let me know if I can help with anything!

[archmoj]

It would be nice to test hoversubplots cases.
@mbant Would you please fetch upstream/master and merge it into this branch?
Thank you!

[archmoj]

Excellent bug fix.
💃