Refactor model compression directory structure

[chicm-ms]

Refactor model compression directory structure as following to better accommodate more automatic pruning algorithms.

compression
  |--tensorflow
  |--torch
       |--pruning (weight masking algorithms)
       |--quantization
       |--controllers? ( TBD,  control the training process, such as automatic compression,  amc)
       |--speedup
       |--utils

Next steps would be:

1. add basic controllers such as one shot , iterative controllers to prune models using all existing algorithms.
2. move algorithm configuration part from pruning package into controller, pruning algorithms can prune specified layers with specified sparsity on demand controlled by controllers.
3. add auto pruners