Bulk-copy byte fields on the read path

[DanielThomas]

HollowObjectTypeReadStateShard.readBytes() currently materializes a BYTES field by calling ByteData.get() for every byte. For segmented on-heap data, each call repeats interface dispatch, segment lookup, and offset masking.

Add ByteData.copyTo(...) and use it to materialize byte fields in bulk:

• SegmentedByteArray copies one contiguous segment range at a time with System.arraycopy.
• ArrayByteData uses System.arraycopy directly.
• EncodedByteBuffer copies chunks from duplicated mapped buffers, preserving thread-safe reads.
• Other ByteData implementations retain a byte-by-byte default.

Benchmark

JMH on Java 8/x86-64. Each invocation includes allocation of the returned byte array. The source begins 17 bytes before a segment boundary, so values of 32 bytes and larger cross at least one boundary.

Field size	Per-byte copy	Bulk copy	Improvement
8 bytes	15.850 ns	12.818 ns	19.1%
32 bytes	38.981 ns	20.132 ns	48.4%
256 bytes	247.829 ns	36.113 ns	85.4%
2,048 bytes	2,023.069 ns	129.415 ns	93.6%

On Java 25 on the same host, the corresponding improvements were 34.3%, 64.7%, 91.9%, and 95.3%.

[DanielThomas]

Gated bulk byte-field copies behind SHARD_READ_FAST_PATHS. Empty/null byte-field fixes remain unconditional.