Figure out logging for multimodal images

[cnmy-ro]

The multi-modality toy dataset is added now, but the code isn't yet usable since the current logging cannot handle multimodal domain A images. Find a solution that can be generalized to many other cases including multimodal medical images like PET-CT.

Possible options:

1. Create a custom multi-modality GANs for this case -- Here, the visuals['real_A'] can be split into visuals['real_A1'] and visuals['real_A2'].
2. Modify tracker code to handle this case -- Split visuals['real_A'] while making the 2D grid for visualization within the train and val trackers.

UPDATE: Going with Option-2, after discussion with Ibro and Suraj