Figure out visualisation for comparing multiple unlearning methods across multiple tasks

[Willmish]

Proposed by Andrzej:

• for one task (harmfulness): visualise "in progress" unlearning at checkpoints (same as before)
• for the other we could just eval and visualise at the end of unlearning (depending on time)
  

How to compare the 3 methods on a single plot?

• Sequential, batch, continuos