Reduce garbage created by ByteBuffersDataOutput

When profiling indexing because of LUCENE-9027, I noticed that indexing produces a lot of unnecessary garbage because of ByteBuffersDataOutput, which can easily get fixed:

• reset() is implemented using streams, which apparently create lots of objects
• copyTo has an optimization for the case that ByteBuffers are backed by an array that never gets used because toBufferList makes the buffers read-only, which in-turn disallows access to the array