Second Batch of Optimizations — IO.apply

[alexandru]

Continuing the work started in PR #90, this pull request optimizes IO.apply.

We do so by introducing a special internal state just for it. This is worth it in terms of performance because it avoids extraneous memory allocations and the numbers look good.

Also the loops described in IORunLoop are now no longer described as @tailrec functions, but as while loops. This led to a slight improvement across the board and it's important because it afforded us the IO.apply optimization in its current form.

Here's the numbers and note that version 0.6-964e8d0 here is PR #90:

Benchmark	0.5	0.6-964e8d0	This PR
ShallowBindBenchmark.async	48.609	83.134	87.018
ShallowBindBenchmark.delay	1535.69	3643.028	5452.494
ShallowBindBenchmark.pure	2556.71	6293.489	6512.625
DeepBindBenchmark.async	2.728	340.461	370.715
DeepBindBenchmark.delay	4.401	5442.471	6469.298
DeepBindBenchmark.pure	2790.456	7133.611	7437.164
AttemptBenchmark.errorRaised	4.265	2122.419	2336.17
AttemptBenchmark.happyPath	111.192	2273.447	2479.291
HandleErrorBenchmark.errorRaised	283.353	1934.246	1987.989
HandleErrorBenchmark.happyPath	1120.249	3058.098	3044.987