Training cannot be properly resumed

[GuillaumeBroggi]

Statement of need

Graphorge saves checkpoints but does not implement a complete training resuming functionality (only the model state can be restored at the moment). Being able to completely resume the training would be a nice addition.

Priority

Low (nice to have).

Difficulty

Easy.

To-do list

• Load optimizer.
• Export loss history at each epoch.
• Load loss history.

Assessment

All the proposed changes are completely decoupled from the rest of the code, impact is expected to be minimal.