Extract independent snapshot functions for metrics: network, RL, experiment, statistics etc.

This will dramatically increase reusability and reduce boilerplate code. 

e.g. 
```python
snapshot_nn(net, ['total_parameters', 'trainable_parameters'])
```

Networks:
- Total parameters: trainable parameters, un-trainable parameters
  - Extract the number of such parameters
  - Extract a generator of such parameters
- Gradient norm

Policy:
- Entropy