DispatchedSplitter should be a case class, consistent with other splitters.

[NathanHowell]

This also has the benefit of inheriting from Serialized, which is required for Spark based training and model evaluation.

[avibryant]

You should also change the creation of DispatchedSplitter in Defaults.scala, right?

[NathanHowell]

I've updated it, as well as a few other formatting tweaks.

[NathanHowell]

With case classes the generated apply method on the companion object just calls new, so it wasn't actually needed but is good for consistency.

[avibryant]

Thanks!