Aggregator dependency injection, min/max, and skipNaN

[Jolanrensen]

Helps #961

• Refactored Aggregator, splitting it into 3 parts:
  • Input, can be supplied by a class implementing AggregatorInputHandler, has 2 implementations
  • Aggregation, can be supplied by a class implementing AggregatorAggregationHandler, has 2 implementations
  • Handling multiple columns, can be supplied by a class implementing AggregatorMultipleColumnsHandler, has 3 implementations

Together they can instantiate Aggregator in any combination you like :)

• Added Aggregator.aggregateByOrNull() functions for selecting aggregators, like min/max, but also median/percentile in the future, with the Sequence.bestBy family of internal functions

• added skipNaN option to sum, mean, min, max

• Rewrote min/max with correct types (T : Comparable<T & Any>?)

  • Decided against allowing mixed number types. Allowing this requires each function to get 3 overloads to avoid ambiguity:
    • <T : Comparable<T & Any>?> for normal comparables
    • <T : Number?> for mixed number types
    • <T> ... where T : Number?, T : Comparable<T & Any?> for normal numbers
  • Plus a slight rethink of the selecting aggregator because each min with numbers would then act like a minBy { it.toUnifiedNumber() } as it needs to return unconverted minimal value
  • But if it's valueble enough, we can add it

• TODO remove skipNaN from min, max again; min nor max can ever return NaN, because any number compared to NaN is considered "better"

• TODO More tests?

[Jolanrensen]

Seems our notebook tests now fail because of https://youtrack.jetbrains.com/issue/KTNB-967/IllegalStateException-null-DefinitelyNotNullType-for-T-exception-while-analyzing-expression