Optimize example-based runtime

[zyteka]

Currently, generating example-based explanations with Pyreal is slightly slower than alternatives because, in order to match with the existing class structures, feature descriptions are applied after-the-fact using ExampleBasedExplanation.apply_feature_descriptions. This is slower than the alternative method of applying feature descriptions to the training dataset at fitting time.

To fix this, the similar examples fitting function should check if feature descriptions are enabled and apply them there, and then apply_feature_descriptions should be updated to be a no-op for similar examples explanations.