Core's handling of unicode string packing is, not to put too fine a point on it, wrong.

OutputBuffer's packString(String s) assumes the serialized form of a UTF-8 string has the same number of bytes as the String has code units, which is incorrect for any String containing multi-byte code units.

Since that needs fixing anyhow, the refactoring pulls out the byte[] packing to a separate function from the String encoding, additionally enabling a performance optimization:

Where a fixed String is packed frequently, the transformation to byte[] can be done once and cached, eliding memory allocation and processing on subsequent invocations. This is particularly useful for the StateManager marker, which is packed to buffers with great frequency.