Enable setting NormalizingEstimator options

[artidoro]

In PR #2959 we hid the NormalizingEstimator ColumnOptions. Currently users cannot set advanced parameters for the estimator.

The reason for not doing that was that NormalizingEstimator has 5 ColumnOptions classes and would have required at least that many overloads. So we opted for waiting for the full Options object to be made accessible and make sure that no breaking API change was going to be required.

However, as per further discussion in #2884 and offline we decided that we would not make Options available before v1. We therefore need another solution.

One possible approach would be to add five new extensions to MLContext each presenting all the settings in one of the NormalizingEstiamtor ColumnOptions. The extensions could indicate in their name the type of normalization to perform.

Note that this is not a breaking API change.

[abgoswam]

It seems to expose the advanced settings for each of the 5 normalizing estimators we have {MinMax, MeanVariance, LogMeanVariance, Binning, SupervisedBinning} we need to add 5 APIs to our public surface as shown in PR #3052

Adding these APIs will also enable to scenario @codemzs highlighted, whereby currently users of ML.NET have no way to set the advanced options for the individual normalizing estimators

@shauheen , @TomFinley should we consider adding these for V1.0 ?

[artidoro]

Note that NormalizingEstimator is on the list of transforms specified by @TomFinley in #2884 (comment) where we should enable multi-column mapping. If we do that, we need to double the number of new extensions.

From 5 we would now go to 10 new extensions.

[TomFinley]

Note that NormalizingEstimator is on the list of transforms specified by @TomFinley in #2884 (comment) where we should enable multi-column mapping. If we do that, we need to double the number of new extensions.

From 5 we would now go to 10 new extensions.

I'm honestly less convinced of that than you all are, since you can always concatenate then normalize (and it's probably clearer to do so).