Performance improvements in Chunk#toArray

[mpilquist]

Fixes #977.

Includes a series of optimizations:

• Chunk#toArray now simply uses a while loop instead of using the segment machinery to access each element (not sure why it ever did what it was doing...)
• toArray is overridden on each array backed chunk subtype using .slice operation
• toByteBuffer was added to Chunk.Bytes allowing a zero-copy view of the underlying array

@fiadliel could you take a look at this?

[fiadliel]

The specialized toArray was the same as what I had, and other changes look great! Throughput for the sync NIO calls is now amongst the fastest options; async is slower, but I think reasonably acceptable:

[info] Benchmark                                              Mode  Cnt  Score   Error  Units
[info] LargeStreamBenchmark.commonsIo                         avgt   10  3.467 ± 0.276   s/op
[info] LargeStreamBenchmark.fs2IoStream                       avgt   10  6.050 ± 0.268   s/op
[info] LargeStreamBenchmark.fs2NioAsyncStream                 avgt   10  7.764 ± 0.237   s/op
[info] LargeStreamBenchmark.fs2NioStream                      avgt   10  2.969 ± 0.102   s/op
[info] LargeStreamBenchmark.monixObservable                   avgt   10  4.307 ± 0.161   s/op
[info] LargeStreamBenchmark.monixObservableWithAsyncBoundary  avgt   10  2.833 ± 0.153   s/op

[mpilquist]

Awesome, thanks for benchmarking!