Rewrote LanguageConverter::autoConvert() to make it use preg_match() with an offset...

Rewrote LanguageConverter::autoConvert() to make it use preg_match() with an offset instead of preg_split(). Reduces memory usage for my test case ([[台灣演員列表]]) to a negligible amount. This should eliminate the most common cause of OOMs on Wikimedia. Produces the exact same output for that test case for the zh -> zh-tw, parser tests pass, seems to work.