Parallelization of computing-intensive functions

I do not know if this an issue there is any interest in, but I noticed the library can potentially benefit from parallelizing parts of computing-intensive functions.

I experimented a bit and used OpenMP to simply parallelize the check if a hexagon is contained in a polygon in the polyfill function: https://github.com/nmandery/h3/commit/65afcb16579e833083feba713a89d7423611212a

Results without openmp:

$ ./bin/benchmarkPolyfill 
    -- polyfillSF: 5830.508514 microseconds per iteration (500 iterations)
    -- polyfillAlameda: 7501.964168 microseconds per iteration (500 iterations)
    -- polyfillSouthernExpansion: 325676.916500 microseconds per iteration (10 iterations)
$ ./bin/benchmarkPolyfill 
    -- polyfillSF: 5241.980240 microseconds per iteration (500 iterations)
    -- polyfillAlameda: 7528.802096 microseconds per iteration (500 iterations)
    -- polyfillSouthernExpansion: 326040.845100 microseconds per iteration (10 iterations)

Results with openmp:

$ ./bin/benchmarkPolyfill 
    -- polyfillSF: 2136.395920 microseconds per iteration (500 iterations)
    -- polyfillAlameda: 2469.901036 microseconds per iteration (500 iterations)
    -- polyfillSouthernExpansion: 98592.296500 microseconds per iteration (10 iterations)
$ ./bin/benchmarkPolyfill 
    -- polyfillSF: 2006.864104 microseconds per iteration (500 iterations)
    -- polyfillAlameda: 2428.976026 microseconds per iteration (500 iterations)
    -- polyfillSouthernExpansion: 97370.778000 microseconds per iteration (10 iterations)

These are the specs of the machine I used:

$ grep 'model name' /proc/cpuinfo 
model name	: Intel(R) Xeon(R) CPU E5-2670 0 @ 2.60GHz
model name	: Intel(R) Xeon(R) CPU E5-2670 0 @ 2.60GHz
model name	: Intel(R) Xeon(R) CPU E5-2670 0 @ 2.60GHz
model name	: Intel(R) Xeon(R) CPU E5-2670 0 @ 2.60GHz

I guess the usage of OpenMP should be optional, as I do not know how this would affect the emscripten based Javascript port of h3.

Start with the polyfill function and the check for whether a hexagon is contained in a polygon, then review bin/benchmarkPolyfill and the OpenMP experiment linked in the issue. Compare the benchmark results and consider the impact on the emscripten-based JavaScript port. Done means an agreed, optional parallelization approach with validated performance and compatibility.