Co-Adaptation of Algorithmic and Implementational Innovations in Inference-based Deep Reinforcement Learning