XidImpl implementations can optimize some byte[] allocations

The EmbeddedXid and XidImpl classes show up on profiler for allocating byte[] in a local environment.

1. We should be able to remove the branch byte[] completely as well as its accompanying AtomicInteger and just return an empty byte[] that is cached (ie. Util.EMPTY_BYTE_ARRAY).
2. We can use the byte[] that is provided from the create method directly in XidImpl, which would reduce our allocation to just the 1 byte[24] for embedded or byte[32] for remote.
3. We can return the byte[] directly for getGlobalTransactionId without copying.