Feature: add arg to resume training from a checkpoint

[royerk]

What

New feature. Resume training by passing the path to a checkpoint.

• [x] non-breaking change & backward compatible

---

How

• Add path arg to argparse
• If path is present then load model

---

Why

You name it: resume training after crash, restart before the training went south...

---

Test

"it works on my machine"

---

Notes

• maybe model.set_env() can be handled better

[glmcdona]

Thanks for this feature, appreciate it!