Make typeclass instances private where possible

[rklaehn]

In a lot of places in spire, concrete typeclass instances are public. This increases the surface area of the library needlessly. As far as I can see, it also does not have any performance benefit in the typical use case.

So I would propose changing e.g. this (spire.std.int.scala)

trait IntInstances {
  implicit final val IntBitString = new IntIsBitString
  implicit final val IntAlgebra = new IntAlgebra
  import spire.math.NumberTag._
  implicit final val IntTag = new BuiltinIntTag[Int](0, Int.MinValue, Int.MaxValue)
}

to this

trait IntInstances {
  implicit final val IntBitString: BitString[Int] = new IntIsBitString
  implicit final val IntAlgebra: EuclideanRing[Int] with NRoot[Int] with Order[Int] with Signed[Int] with IsIntegral[Int] = new IntAlgebra
  import spire.math.NumberTag._
  implicit final val IntTag : NumberTag[Int] = new BuiltinIntTag[Int](0, Int.MinValue, Int.MaxValue)
}

This would allow making everything in spire.std.int.scala except IntInstances private or at least private[spire]. Doing the same everywhere where typeclass instances are defined would result in a large reduction in surface area.

However, this would break some code in case somebody uses e.g. IntOrder in another namespace. So I am not completely sure if this is a good idea. @non @tixxit what do you think?

[tixxit]

I think I tried this before (making things private) and may have hit some issues... possibly with specialization? I can't quite remember what the hang up was - so just a warning that there may be some (waves hands) issues (/waves hands).

[rklaehn]

OK. Thanks for the warning. How would I notice if there are ... issues ... ? Are there any benchmarks that would detect completely screwing up the performance of typeclass based operations?

[tixxit]

These were compiler issued warnings or errors, I think. Nothing too subtle. So if all is go in well, then its probably ok.