fix surrogate pair byte counting in ExtendedBufferedReader array read

[digi-scrypt]

1. with byte tracking on (setTrackBytes + a charset), read(char[]) advances lastChar to the last buffer char before counting, and the per-char helper reads that field instead of the actual preceding char, so a surrogate pair gets matched against the wrong neighbor (the loop also ran to length instead of offset+length).
2. a 4-byte char taken through the char[] path, e.g. a multi-character delimiter holding a supplementary character, then throws CharacterCodingException out of nextRecord and getBytePosition() goes wrong.

Passed the previous char explicitly and moved the counting ahead of the lastChar update. What happens for a pair split across two buffer reads is covered too since the first char still pairs against the saved lastChar. Added a regression test next to the existing multi-character-delimiter byte-position one.