Lookahead Anchoring: Preserving Character Identity in Audio-Driven Human Animation