make offsetgroup work with barmode 'stacked' and 'relative' for bar traces

[my-tien]

This change makes it possible to create groups of stacked bar traces by setting barmode: "stacked" (or relative) on the axis and setting the same bar.offsetgroup for bars that should stack together. If no offsetgroup is defined, all bars at the same position are stacked together.

To summarize the interaction between axis.barmode and bar.offsetgroup after this change:

barmode 'group': bars without offsetgroup are automatically offsetted to appear next to each other. If some bars should appear at the same location, they should have the same offsetgroup.

other barmodes: bars without offsetgroup share the same position on the axis. If some bars should appear next to each other, they should have differing offsetgroups.

[archmoj]

@my-tien the mapbox baseline failures are fixed in #7019.
Please fetch upstream/master and then merge master into this pull request as well as into other PRs.

[archmoj]

Getting closer.
Could you please investigate why the legendgroup mock renders differently compared to master?

[my-tien]

Thanks @archmoj, I was able to restore the original looks of the legendgroup mock.

Two general questions to ensure that my changes make sense:

1. I am a little bit unsure about the intended interaction between different trace types and their bar/scatter/box...modes. As far as I understand it, across traces we have the implicit behavior of "overlay" (with the exception of bars and histograms, because histograms also use barmode). Is this correct?

2. How should alignmentgroup work exactly? The description in the documentation wasn't clear to me.
   And should alignmentgroups only matter among traces of the same type? I was experimenting with bars and boxplots in this codepen and saw some interaction. Boxplots become invisible when a bar shares its alignmentgroup:
   https://codepen.io/toffi-fee/pen/zYVaKrB

EDIT: I accidentally overwrote the content of the previously linked codepen with something else. Now I've recreated my example:

And here is the output of the same figure in my branch:

[alexcjohnson]

@my-tien yes, that’s all correct, and thank you for catching my error, I will fix it above.

[my-tien]

After spending way too much time trying to implement the implicit alignment and offsetgroups I figured that I don't actually need that for this PR.
As long as alignmentgroup and offsetgroup are set explicitly for all traces, they work as expected.

Regarding the two still failing baseline tests, the new images look to me more like what I would expect based on their mocks (ticks and grid at whole numbers). What do you think?

If you agree, I'll replace the baseline images with the new ones.

E.g. compare current baseline:

New image:

[archmoj]

@my-tien Would you please fetch upstream/master and merge it into this branch?

[my-tien]

done @archmoj

[archmoj]

@my-tien Would you please fetch upstream/master and merge it into this branch again?
We are thinking to possibly include this feature in v3.1.0.
Thank you!

[archmoj]

Thinking about this PR, it looks like a change that could possibly include in v3.0.0. @gvwilson
@my-tien Would you please merge master into this branch so that the tests pass?
Also please add draft log with a _change suffix.
Thank you!

[stephprobst]

@archmoj: I opened #7247 and rebased the branch on the latest master (I don't have access to my-tien's original branch). If you want, you can merge the new PR #7247 and close this one. I also added a draft log with _change suffix as requested.

[archmoj]

@archmoj: I opened #7247 and rebased the branch on the latest master (I don't have access to my-tien's original branch). If you want, you can merge the new PR #7247 and close this one. I also added a draft log with _change suffix as requested.

Thanks @stephprobst I'm working on the same thing right now.

[archmoj]

IMHO this "change" would be best to go into v3.0.0 because that could be considered a breaking change.
@alexcjohnson @gvwilson Do you agree?

[gvwilson]

👍 on including this in 3.0 cc @LiamConnors we'll have to document the changed behavior

[my-tien]

Many thanks for wrapping this up in my absence! <3