Performance: Reduce synchronization

[benjchristensen]

Performance changes as part of #104 and is an extension of work done in pull #106.

This code was used to test the performance:

    public void runCompositionTest(Integer[] values) {
        System.out.println("runCompositionTest");

        final AtomicInteger onNextSum = new AtomicInteger(0);
        final long start = System.nanoTime();

        MathFunction m = new MathFunction();
        // 50 levels of composition (same function so that's not the cost)
        Observable.from(values)
                .map(m).map(m).map(m).map(m).map(m).map(m).map(m).map(m).map(m).map(m)
                .map(m).map(m).map(m).map(m).map(m).map(m).map(m).map(m).map(m).map(m)
                .map(m).map(m).map(m).map(m).map(m).map(m).map(m).map(m).map(m).map(m)
                .map(m).map(m).map(m).map(m).map(m).map(m).map(m).map(m).map(m).map(m)
                .map(m).map(m).map(m).map(m).map(m).map(m).map(m).map(m).map(m).map(m)
                .subscribe(new TestObserver(onNextSum, start));
    }

Before it took:

Sum: 710082754 Time: 60.605ms

By reducing the nested synchronization (eliminating most of it) it is now running at:

Sum: 710082754 Time: 22.429ms