fix oob read in UTF-16 byte-swap LA from wrong nextChar pointer type

[zayeem06]

antlr3UTF16LALE and antlr3UTF16LABE declare their scan pointer nextChar as pANTLR3_UCHAR, which is a UINT32*, but the code assembles each character a byte at a time and steps with nextChar += 2, so every access reads four bytes and strides in four-byte units. A well-formed UTF-16 source whose byte order is opposite to the host (for example a UTF-16BE file on a little-endian build, which selects these byte-swap routines) then reads past the end of the input buffer and hands back wrong code points; ASAN reports a heap over-read at the character fetch. Declaring nextChar as pANTLR3_UINT8 makes the byte arithmetic match the reads.