Linear SVM lambda parameter name should be public

[gregorybchris]

Describe the bug
The Linear SVM learner was recently added to NimbusML (version 1.3.0) and the parameter lambda, which is available as an optional regularization constant (called Lambda) for sweeping in ML.NET, is named lambda_ in NimbusML. This seems to imply that it is not part of the learner's public interface (or it was a typo).

To Reproduce
Steps to reproduce the behavior:

1. Pass either lambda or lambda_ to the linear_model.LinearSvmBinaryClassifier constructor.

Expected behavior
The lambda parameter should be used to make it clear that sweeping this value is allowed from the public interface.

Enabling lambda as an optional parameter will be backward compatible since lambda_ should not have been used by any consumers of package versions 1.3.0 and 1.4.0.

[gregorybchris]

I just had a sudden realization that lambda_ is used in favor of lambda due to Python reserved word restrictions. Now I wonder if it makes sense to update the parameter name to be something like regularization or regularization_coefficient. This might be confusing for people who use both NimbusML and ML.NET, but it also might be more intuitive for Python users.