Make TextFeaturizer's extractors etc. configurable again

[Zruty0]

#801 is making the parameters default and hardcoded for the following options of Text featurizer:

• Stop word remover (defaults to none) - Fixed in Made 'StopWordsRemover' in TextFeaturizer configurable again. #2962
• Custom term dictionary (defaults to none) - This parameter is hidden
• Word feature extractor (defaults to unigrams) - Fixed in Exposed ngram extraction options in TextFeaturizer #2911
• Char feature extractor (defaults to 3-char) - Fixed in Exposed ngram extraction options in TextFeaturizer #2911

Once individual building blocks become estimators, we should bring these parameters back (in a form of estimator for word/char extractor etc.).

Or maybe we shouldn't, and instead just demonstrate how to compose your version of text transform from the individual building blocks?

[justinormont]

The text transform is an incredible time savings and lowers the bar to entry for making a good NLP model.

Building from the individual blocks is a rough road to travel, and doesn't add much extra power. The only case I recall needing to use the individual blocks was when using the lemmatizer, which I don't think is available in ML.NET.

Also, I'd recommend Bigrams+Trichar as the defaults, which matches our default text recipe.

[eerhardt]

Is this strictly adding new API? Can this be done without a public API breaking change? If so, I think we can remove it from Project 13, and it can be added after v1.0.

But if this requires a public API breaking change, then it can be left in Project 13.