[FEATURE] Add Video Logging during training

[jselvaraaj]

It would be really nice to log videos of agents' behavior during and after training to TensorBoard.

Proposed Solution:

1. Add logic to the existing Mava/Tensorboard logger class to handle the collection of RGB frames.
2. Call this new video logging function from the logger class during the evaluation phase of the algorithm.

[sash-a]

We've wanted this for a while and would be a great addition. The problem is that some envs render differently to others so we'd need to unify the API to the jumanji style (env.animate(states)) which is why we haven't yet done this.

Doing it during eval is a good idea, but also a bit of work, you'd need to return all the states from the eval loop and then filter them for a single episode and then render that.

I think the idea of adding an animate/render method to the loggers is great 🔥