Improve lock performance

[kmcbride]

Replaces existing NSRecursiveLock and serial DispatchQueue synchronization with unfair locks. This isn't using OSAllocatedUnfairLock due to the supported deployment targets, but it's something to consider in the future.

Benchmark	Before	After	Speedup
Lock.synchronized (100k)	0.013s	0.011s	1.2x
Synchronized read (100k)	0.044s	0.014s	3.1x
Synchronized write (100k)	0.029s	0.015s	1.9x
Synchronized mutate (100k)	0.043s	0.014s	3.1x
Synchronized compareAndSwap (100k)	0.044s	0.014s	3.1x
Contended mutate (4 threads, 10k each)	0.133s	0.009s	14.8x
Contended mixed R/W (4 threads, 5k each)	0.142s	0.015s	9.5x
EffectExecutor emit (10k)	0.037s	0.021s	1.8x
EffectExecutor concurrent (4 threads, 2.5k)	0.042s	0.028s	1.5x
ThreadSafeConnectable accept (10k)	0.002s	0.002s	~same
ThreadSafeConnectable concurrent (4 threads)	0.009s	0.006s	1.5x

There's also an additional tweak with a much more modest (~10%) improvement for resolving route handlers.