CustomStopWordsRemoverTransform shouldn't hard-code its column name

[eerhardt]

See the conversation here:
https://github.com/dotnet/machinelearning/pull/700/files/e30785ff454627c7b74e3b6873d70b9e6de28bdf#r212152727

We should consider making the column name configurable instead of hard-coding the column name, which may be hiding an existing column.

/cc @codemzs

[Zruty0]

Frankly, this way of loading stopwords doesn't even make sense to me from the API perspective.
The transform allows FOUR separate ways of providing stopwords:

• a string array
• a comma-separated string
• a column in the training data
• a 'Stopwords' column in the separate file, loaded by the provided loader.

I believe we should trim it down to ONE: a string array.

[Ivanidzo4ka]

Since I plan to work on conversion all text transform component to estimators, I'll assign this to myself, I need to touch it anyway.