Improve evaluation metrics and pipeline

[afspies]

This PR Will contain:

• Consolidation of metrics for evaluating models
• Helpers for executing and logging the full suite of evaluations
• Restructuring of plotting scripts
• Changes to plotting function outputs (figures, axes etc.) for unified interface with wandb logging

This PR will NOT contain (required for proper integration):

• Integration of evaluation pipeline into current training script (awaiting further pushes from @rusheb )
• Proper command-line based overriding of config parameters required for sweeping and partially accounted for in evaluation pipeline (I will do this manually in my own training sweep script for the time being)

All contents of maze_transformer/evaluation will be affected. Only minor changes to imports and usage of plot.show are to be expected in maze_transformer/evaluation/eval_model.py and notebooks/eval_model.ipynb

[mivanit]

stale, some evals rework will happen in #207