perf: Improved decoding from InputStreams

[bogovicj]

This PR improves decoding performance by using the known decoded size of a block in bytes. The primary changes are:

• Calling limit on the ReadData used in FlatArrayCodec
• Overriding limit in InputStreamReadData so that it can have an expected size without materializing
  • The default limit method calls slice. and InputStreamReadData.slice calls materialize.

We call org.apache.commons.io.IOUtils.toByteArray(...) to materialize InputStreamReadData
of unknown size, and have previously identified that this call is much slower than the alternative path with known size.

An InputStreamReadData can arise during de-coding, for example during ZStandard decompression.

See also #161
@tpietzsch @cmhulbert