Motivation:
The setBytes method was getting a sliced nioBuffer of its source, which
typically causes allocation. On Java 16 onwards, we can instead copy
using an absolutely offsetted `put` method, and forego allocating a
duplicate ByteBuffer instance that is otherwise needed for isolating the
position field.

Modification:
- Make use of the absolutely offsetted put method in setBytes, when its
available.
- Use the underlying ByteBuffer of the shared chunk where possible to
avoid multiple bounds checks.
- Add a benchmark targeting the setBytes method that takes a ByteBuf
source.
- Change a few benchmarks to use the default allocator when pooling is
enabled.

Result:
Faster setBytes in certain cases.