training script migration

[alphalm4]

omni-train branch
@Jaesun0912

• L2MAE
• OrderedSampler, Batch training
• L2 regularization loss
• train_shift, train_scale
• Grad clip
• things related to large-dataset training (lmdb, dataset.py)
• Acceleration option in input.yaml already implemented in sevenn train cli
• Test scripts for aboves

[Jaesun0912]

TODO: PR