Active Learning with Selective Time-Step Acquisition for PDEs