Time series improvements

[montebhoover]

Mainly for @codemzs, I wanted to ask what additional work is needed for time series. Here are few things I noticed:

1. Add support for doubles in SSA and IID estimators. Currently we check to verify floats, for all other types we throw an exception.
2. Create class with named properties for output metrics of Martingale, P-Value, Alert, and Score. Currently we output a column of vector type that contains these four values. Alternate approach would be to have four output columns.
3. Move everything under namespace Microsoft.ML.Runtime.TimeSeriesProcessing into Microsoft.ML.TimeSeries. Is there a reason why we need both namespaces?
4. Add missing unit tests. We are missing IID and SSA spike detection, and we don't have code coverage over types such double, int, etc.
5. Improve naming for estimator arguments. Currently we have Source for our input column and Name for our output column. It would make more sense to be Source and Dest or InputColumn and OutputColumn.
6. Format the samples for the docs site. For each transformer we have the option of using the transformer on a batch of data or creating a prediction engine - which should we be displaying as the sample on the docs site? Both on separate pages?

[montebhoover]

7. Publish documentation on the technical details of our SSA and IID implementations. This documentation currently exists internally (and is pretty high-quality). What is the ideal way to present this technical documentation? Blog post?

[montebhoover]

8. Update SSA and IID Estimators to be accessed through MLContext. I know there is an open issue for this work TrainerEstimators in particular, but is there an issue open to capture this work?