[FEEDBACK WANTED] Don't create column from dataset

[frosforever]

I may be missing some context here, but it seems like the current way of selecting a column can lead to runtime errors that the compiler okays.
Currently, columns are selected using the dataset's col method. The compile time check however only ensures that the field exists in the Dataset's type T. However, this does not prevent one from selecting a column from a different dataset of the same T at compile time resulting in a runtime blow up.

One way to get around this is started in this PR. Instead of selecting the dataset's col, create a general column instead. This would at least line up with what the compiler is proving, simply that a col of field foo exists in T and has the type U but not that the column requested is that specific dataset's column.

example:

val ds1: TypedDataset[X2[Int, Int]] = TypedDataset.create(Seq.empty[X2[Int, Int]])
val ds2: TypedDataset[X2[Int, Int]] = TypedDataset.create(Seq.empty[X2[Int, Int]])

//this will compile but blow up at runtime
ds2.filter(ds1('a) === 1)

Are there other ways to get around this?
Is this not considered an issue to be concerned about?

[frosforever]

Never mind! Seems like #186, #187, #110 hit on this exact issue. My mistake for not seeing that earlier.