This (native Java) implementation appears to be about as fast as
the old (JNI) implementation, but has better support across more platforms.
Going to evaluate stability on this branch for possible later merging into master.
Breaks only for users that were overriding `freeze-to-bytes` or `thaw-from-bytes` defaults to disable
compression.
Changed to map-based config to allow extra control (like :read-eval? as suggested by Luc Préfontaine).
This change will also allow further additions in future without the need to break the API again.
Luc Préfontaine kindly pointed out that `.writeUTF` and `.readUTF` have a maximum length of
64k. This commit switches to a custom integer-length prefix for strings and reader types to
boost their max length. Keywords are unchanged.
Old string and reader type IDs are now deprecated but will continue to be supported for
thawing.
`write-as-bytes!` was unnecessarily general. Renamed as type-hinted `write-biginteger!`.
Improves write performance of all types that relied on `write-as-bytes`: bigints, bigdecs,
rational.
