Tutorial for Robust Gaussian Processes via Relevance Pursuit (pytorch…

…#2707)

Summary:
This commit adds a tutorial of the robust Relevance Pursuit model, specifically using the `RobustRelevancePursuitSingleTaskGP` and the Twitter flash crash examples, as well as a synthetic regression example that showcases both usage patterns, as well as ways of analyzing the results of the model.


Test Plan:
Ran the tutorial on an internal devserver, my local machine, and the smoke test passes in the [OSS test](https://github.com/pytorch/botorch/actions/runs/13021130751/job/36321841195?pr=2707):

```
Running tutorial relevance_pursuit_robust_regression.ipynb.
Running tutorial relevance_pursuit_robust_regression.ipynb took 7.77 seconds. Memory usage started at 47.5 MB and the maximum was 798.1875 MB.
```

## Twitter Flash Crash Example
{F1974794186}{F1974794179}

Reviewed By: Balandat

Differential Revision: D68353581

Pulled By: SebastianAment

botorch/models/robust_relevance_pursuit_model.py

@@ -61,7 +61,6 @@
 from gpytorch.module import Module
 from torch import Tensor
 
-
 # default fractions of outliers to consider during relevance pursuit
 FRACTIONS_OF_OUTLIERS = [
     0.0,
@@ -117,8 +116,8 @@ def __init__(
             dim=dim,
             convex_parameterization=convex_parameterization,
         )
-        self.bmc_support_size: Tensor | None = None
-        self.bmc_probability: Tensor | None = None
+        self.bmc_support_sizes: Tensor | None = None
+        self.bmc_probabilities: Tensor | None = None
         self.cache_model_trace = cache_model_trace
         self.model_trace: list[SingleTaskGP] | None = None
         self.prior_mean_of_support: float = (