My research indicates they should be treated as Big-Endian.
This leaves UTF-7 as the only encoding known to be unsupported by CsvChunk; it will be handled by parent.
Chunk routine would fail if chunk ended between two halves of a surrogate. Fix that.
UTF-7, UTF-16 and UCS-2 (no endian-ness specified and no BOM), UTF-32 and UCS-4 (no endian-ness specified regardless of BOM) all revert to parent routine.
CsvChunk will not be moved to production, so its test should not be in the regular unit test suite. However, it still has some value in the Benchmark suite, so move it there.
Drop the clone-based copy() since benchmark results vary across PHP
versions and platforms. Keep the benchmark test (renamed to end in Test)
in a separate testsuite so users can assess which approach suits their
environment. Add docblock @see references on copy() and __clone().
oleibman
kemo