fix out-of-bounds read in antlr3UTF8LA backward traversal

[zayeem06]

antlr3UTF8LA walks backwards over UTF-8 continuation bytes to find the start of a code point when asked for a negative lookahead, but the inner skip loop had no lower bound. On input that begins with continuation bytes (malformed UTF-8), a consume followed by LA(-1) steps the pointer before input->data and reads input->data[-1], a heap buffer under-read that AddressSanitizer flags at the skip loop. Bounded the scan with nextChar > input->data so it stops at the start of the buffer; well-formed input is unaffected because the first byte of the buffer is never a continuation byte.