Engression

Update `StoNet` with residual blocks.

Updates:
- prediction and sampling with adaptive batch sizes to avoid out-of-memory error
- standardization for multivariate Y
- fix torch numerical issue when computing the energy score with a float beta

sigmoid` option for classification or response values with support [0,1]^d.

`resblock` defaults to False.

Two new options:
- Classification. Set `classification=True` in `engression` function.
- Residual blocks. Set `resblock=True`. Defaults to True when num_layers > 2.

minor updates on data standardization

add energy distance