Correlation stats and improvements to logger

[nikita-0209]

1. add correlation statistics: correlation can be calculated in two ways: a. ratio of the true and predicted density (for small sample spaces like a 10x10 grid) (type: density_ratio) OR b. sampling all possible trajectories to states in the test set. (type: from_trajectory) The way correlation should be calculated is specified as a member variable of the environment (can be specified elsewhere as well, open to discussion), corr_type. If not specified, corr=0.0.
2. frequency of logging plots is now different from the frequency of testing (hence, variable do_plot is introduced)
3. add an option to resume a wandb run given a run_id
4. compute buffer train and test metrics
5. adapt gflownet.py to work with list of tensors