DiceSGD

Existing codes:

1. ClippedSGD, Clipped-EFSGD, DPSGD, DiceSGD
2. Run python train.py --algo DPSGD/DiceSGD to train vit_small on Cifar-10

Todos:

1. update the calculation of DP noise multipliers for DiceSGD
2. Add an ADAM version for DPSGD and DiceSGD
3. Add new datasets and models
4. Add code for loading pretrained models