When to Sense and Control? A Time-adaptive Approach for Continuous-Time RL