Co-learning Planning and Control Policy using Differentiable Formal Task Constraints