Fix elasticY with evadeDomainFilter

[dahlbyk]

Fixes that evadeDomainFilter, introduced by 77d22ba for #949, doesn't work with elasticY(true). The fix requires access to the domain-filtered values separate from all values needed to render the chart.

Bonus fix: x().domain() transitions from [0, _] to [x, _], as in the bar transitions demo with "right" selected, will now rescale() correctly.

[gordonwoodhull]

Thank you for all these excellent contributions, and sorry for my slow response.

I was in a car accident last week and broke my collarbone, so I haven't been able to spend much time at the computer.

I am very excited to review your three PRs when I'm able, but it won't be for at least another week.

[dahlbyk]

I was in a car accident last week and broke my collarbone, so I haven't been able to spend much time at the computer.

Oh my, so sorry to hear that. Hope you have a quick recovery!

[dahlbyk]

Rebased onto 3.0.

I'll leave the 2.x version at da3c437...NewBoCo:elasticY-evadeDomainFilter-2.x in case anyone needs it.

[gordonwoodhull]

Ok, I was about to look at d3.stack #1375 so I figured I should merge this first.

Thanks @dahlbyk, this makes a lot of sense. I added a simple test, too - turns out we already covered this situation but didn't test it.

Let's fix those line endings too.

I'm dropping the fix to compareDomains because @kum-deepak has a more comprehensive fix in the pipeline in #1402.

This will go into 3.0 alpha 7.

[dahlbyk]

#1391 (comment): I found it a bit confusing but maybe I didn't stare at it long enough.

To clarify what I'm trying to fix, with elasticY(true) added to the bar-transition demo:

With evadeDomainFilter(false)

Note bars outside the filtered domain transition up/down.

With evadeDomainFilter(true), broken

Bars outside the domain filter now slide in/out (the point of evadeDomainFilter), but the y-axis is not elastic because flattenStack always looks at values for the full domain.

With evadeDomainFilter(true), fixed

Now y-axis is elastic because flattenStack considers values for the current domain (saved in layer.domainValues) while layer.values keeps all values to evades the domain filter.

[gordonwoodhull]

Thanks for the demo! That's very helpful for documenting the change going forward.

Yesterday when I looked at this PR again, I understood it, and this fix is already on the 3.0 branch along with a test (linked above).

I haven't tagged it yet. Hoping to include it in alpha 7 with part of #1402 today.