Yet another Bucket refactor!?

[jfirebaugh]

Yeah, I know, I've been refactoring Bucket for years. But it needs it!

This series of refactors culminates in the elimination of ArrayGroup and BufferGroup. These classes were a poor abstraction: they required highly conditional logic to satisfy the needs of all their consumers, and they were tightly coupled to each other. It's better to have each bucket class contain exactly the array/buffer properties it needs, and to keep corresponding arrays and buffers close to each other in code.

This brings the bucket classes much closer to their corresponding gl-native implementations.

benchmark	master 4eb5d98	bucket-interface 7998c6d
map-load	121 ms	111 ms
style-load	64 ms	58 ms
buffer	983 ms	976 ms
fps	59 fps	60 fps
frame-duration	5.8 ms, 0% > 16ms	5.9 ms, 0% > 16ms
query-point	0.76 ms	0.73 ms
query-box	41.95 ms	42.29 ms
geojson-setdata-small	3 ms	2 ms
geojson-setdata-large	93 ms	109 ms

[mourner]

Weird, the Circle build log keeps stuck in loading for me...

[jfirebaugh]

In debugging the query test failure, I think I found a flow soundness bug! facebook/flow#4580

[anandthakker]

😍 Eliminating the `{Array,Buffer}Group abstractions makes the bucket code more explicit and much clearer.

Only one minor question: would it be worth extracting into a initializeBucket() helper the bucket initialization code that's repeated in all but the symbol bucket (and actually, there too, in SymbolBuffers)?

[jfirebaugh]

There's definitely some commonality that deserves extraction, but I want to avoid creating an abstraction until it's clearer that we've got the right one. Some of the issues are:

• SymbolBucket is an obvious outlier.
• FillBucket is also an outlier -- it maintains two index buffers per vertex buffer.
• I think we'll want to make the flow types involved here generic on the layout attributes, like they are in native.

@kkaefer had some good ideas in this vein in mapbox/mapbox-gl-native@2d36983; see my comments on mapbox/mapbox-gl-native#9468. (These changes were written out of that PR but I've pushed a branch to keep them around.)