Vertex optimizations, first pass

[nrabinowitz]

Small optimizations for cellToVertex. The main benefit of this work was to identify the key perf bottleneck, which is the call to _h3ToFaceIjk in vertexRotations. The optimizations here are mainly to avoid expensive calls:

• Skip calculating left and right neighbors if the origin is a center child (center children will always have the lowest index)
• Skip calculating right neighbor if the left neighbor is a center child
• For hexagon cells, use a simple direction lookup table to determine the reverse direction, instead of using the slightly slower directionForNeighbor.

This improves things, but doesn't add up as much as I'd like - approx 20% speedup:

Before

	-- cellToVertexes: 257.191600 microseconds per iteration (10000 iterations)
	-- cellToVertexesPent: 267.356200 microseconds per iteration (10000 iterations)

After

	-- cellToVertexes: 13.119300 microseconds per iteration (10000 iterations)
	-- cellToVertexesPent: 0.357300 microseconds per iteration (10000 iterations)
	-- cellToVertexesRing: 215.354700 microseconds per iteration (10000 iterations)
	-- cellToVertexesRingPent: 227.581100 microseconds per iteration (10000 iterations)

I think I can improve on this significantly by refactoring cellToVertexes to avoid repeated calculations for the origin and neighbors, but that will come in a later PR.

[coveralls]

Coverage increased (+0.07%) to 99.25% when pulling 9038f7d on nrabinowitz:vertex-optimizations into dd0b1e5 on uber:master.

[dfellis]

Looks good. The overall complexity isn't much higher and the perf bump is good.