CLARE: Continual Learning for Vision-Language-Action Models via Autonomous Adapter Routing and Expansion