Meta-BGD strategy

[HamedHemati]

I'm creating this issue to implement the Meta-BGD strategy from this paper: Task Agnostic Continual Learning via Meta Learning

It's a commonly used baseline in meta-learning-based strategies.