[enhancement] Decouple Regression Model validation from building a regression model

### What's hard to do? (limit 100 words)

Currently, we do k-fold validation every time we call the constructor [1] of RegressionEstimator. The k-fold validation is private within the class. However, there's a need for using the validator without using the model [2], which calls the EstimatorModel to do k-fold validation. This is confusing.

[1] https://github.com/google/xls/blob/6497966ee820e29c7b86552511aaddb2a54430af/xls/estimators/delay_model/delay_model.py#L355-L360
[2] https://github.com/google/xls/blob/6497966ee820e29c7b86552511aaddb2a54430af/xls/contrib/integrator/area_model/area_characterization_client_main.py#L405-L406

### Current best alternative workaround (limit 100 words)

N/A

### Your view of the "best case XLS enhancement" (limit 100 words)

Decouple RegressionEstimator validation from constructor, allowing validation without using the model.