Fix TableDisplay column order for Spark Datasets

[jpallas]

Hidden in the example of #7393 is a problem I noticed a while ago but forgot about. The column ordering gets scrambled because the default Map doesn't preserving order for more than four columns (it does up to four, presumably an optimization for small maps).

Two approaches:

• Use ListMap. Minimizes changes to existing code.
• Use the TableDisplay (values, columns, classes) interface instead of building all those maps. Classes can be derived from the Dataset schema.

I'm leaning toward the second. I don't think the table display guesses some of the types correctly currently.

[scottdraves]

great point, thank you.