Only store Sigma values in BART samples instead of object (#1527)

Summary:
Pull Request resolved: #1527

Background:
We are building Bayesian Additive Regression Trees (BART) as an experimental causal inference model in beanmachine. Details of the project can be found in https://docs.google.com/document/d/11nkB6UTGpvQBEC2yBjfgwAr8VabTlD7R9XufGQG0EvI/edit?usp=sharing and the proposed design can be found in the draft design document: https://docs.google.com/document/d/1o3J7yobDF0M9E27Y0tP2889fycmemXUZbHE5cebRqzs/edit?usp=sharing.

In this diff:
The noise standard deviation (sigma) parameter is never really used in the prediction tasks. While we would like to retain them for diagnostic purposes, there is no reason to store the NoiseStandardDeviation object in the sample trace. In this diff, we are modifying the BART class to only store float samples of the noise standard deviation.

Reviewed By: feynmanliang

Differential Revision: D37635208

fbshipit-source-id: be2f53b61b666fe9d50d2504a57351c91bd24915

src/beanmachine/ppl/experimental/causal_inference/models/bart/bart_model.py

@@ -219,7 +219,7 @@ def _step(self) -> Tuple[List, float]:
                 self.X
             )
         self._update_sigma(self.y - self._predict_step())
-        return self.all_trees, self.sigma
+        return self.all_trees, self.sigma.val
 
     def _update_leaf_mean(self, tree: Tree, partial_residual: torch.Tensor):
         """