Residual Skill Policies: Learning an Adaptable Skill-based Action Space for Reinforcement Learning for Robotics