Fix hover label exponents

[rreusser]

This PR fixes #755 and #1711. After a bit of digging:

The problem: Hover labels are sometimes wrong by arbitrarily gigantic orders of magnitude, which is bad. Hover labels piggyback on Axes.tickText, which sometimes hides exponents and just shows 1/2/5 values. That's always invalid for hover labels since their exponent is not shown elsewhere as it is for axes.

The solution: whenever 'hover' is passed to tickText, this PR prevents it from making any automated decision about whether to hide the exponent, instead just obeying the axis formatting.

[alexcjohnson]

Looks good to me. I could imagine with showexponent: 'none' wanting the hover labels to have no exponents either - but that's a weird case and I could equally imagine wanting the exponents only on hover in that case. If you want that kind of control you can make your own hover text :)

This whole system of formatLinear/formatLog/numFormat is a bit of a palimpsest at this point and probably should be completely refactored... but for now this is great.

I took the liberty of merging #1930 first, so you can fix these tests 🎉 and then 💃

[rreusser]

Haha good word. Yeah, the logic was about a 6/10 for subtle and intricate, so I was a bit worried I'd miss some cases. As is often the case then, I focused mainly on fixing the bug while adding and modifying as little as possible. I don't quite understand the above comment about showexponent: 'none' (like 0.0000000000000000112 then?) but seems like the top priority is that they're not wrong. Opting for merge. 😄

[etpinard]

Thanks for the 📚 @rreusser

[etpinard]

Very nice PR @rreusser thanks for doing this 🎉

🔪'ing two issues in ~10 new LOCs is amazing!

[etpinard]

@rreusser this PR breaks this gl3d hover label test. CI ✅ as this particular test is skipped on CI.

Can you make sure npm run test-jasmine -- gl_plot_interact passes?

[rreusser]

Sure. En route to meeting with Alex today so will fix during the flight, along with the final trace naming issue.