Improve buffering in Compression Streams API

[fhanau]

• Eliminate recursion in compression streams writeInternal()
  This avoids having large stack traces when compressing larger chunks of data
  and is easier to maintain with the added comments.

• Reduce excessive copying in Compression Streams API
  In the Compression Streams API, introduce a helper class to only shift output
  data when this would reduce output vector size significantly. This results in
  the leading bytes of the output vector becoming invalid when reading only parts
  of it, which requires careful tracking but avoids having to move the output data
  with every read from the buffer.

• Increase Compression Streams output buffer size
  Since zlib uses 256KiB of memory per compression stream with default settings, a
  size of 16KiB will not increase memory usage by a large proportion. This should
  make processing data slightly faster. Ideally we'd want this to be larger than
  deflate's 32KiB window size which would allow zlib to avoid some extra copying
  during decompression, but that can be done at a later time alongside other
  compression optimizations.

• Clean up ArrayPtr casts