DäRF: Boosting Radiance Fields from Sparse Input Views with Monocular Depth Adaptation