[core] port the pattern-fill antialiasing fix from gl-js

[mollymerp]

just like this

[mollymerp]

Render tests from Travis are here: https://mapbox.s3.amazonaws.com/mapbox-gl-native/render-tests/11966.1/index.html

It looks like this change is affecting more of the render tests than it should be. Render tests that I thought were being affected by this PR are actually consistent with the most recently merged PR build. The condition I'm using to see whether to apply the outline-pattern shader is if (properties.antialias && stroke_color == fill_color) {}.

@jfirebaugh mentioned in chat that I should try the conditional stroke_color[3] == -1 to test if the stroke-color property of a fill layer is not set, and if it isn't, apply the outlinepattern shader. That didn't seem to work (i.e. there was no change from the master branch in the rendering tests)

I will continue poking around, but if anyone has suggestions I'd love to hear them!

[jfirebaugh]

@ansis Can you help @mollymerp here? 🙇

[ansis]

The render tests look right to me. The antialiasing doesn't match -js but that's expected because -js renders antialiasing wrong in headless testing: mapbox/mapbox-gl-js#2080

I think the failing outline test can be disabled in -native until we fix the antialising issue.

[ansis]

I think this should be setDepthSublayer(2); like the non-pattern antialiasing call.

What this does is draws the outline at a further depth so that it doesn't get drawn on top of the inner part of the fill.

[ansis]

Once https://github.com/mapbox/mapbox-gl-native/pull/4771/files#r61337105 is fixed and tests failing because of antialiasing differences are disabled I think this should be good to merge.

@jfirebaugh do you still want to review anything here?

[jfirebaugh]

LGTM!

[mollymerp]

Updated the render test for test-Map.Offline that was failing:

old expected

new expected

thank you to @tmpsantos for helping me troubleshoot the failing builds! 💯

once everything is green, this is good to 🚢 @ansis ?

[ansis]

🚢!