Unified number types

[Jolanrensen]

Fixes #1068
Helps #961

Fixed getCommonNumberType and commonNumberClass functions that are used solely by sum at the moment.

I introduced a future-proof rewrite of the function and added support for unsigned- and big numbers. Moved it to a separate file and added tests. We will reuse this logic in more places later. I created a small DAG implementation for this, as it's smaller than yet another dependency.

           BigDecimal
           /       \
     BigInteger     \
       /    \        \
   ULong   Long    Double
..   |    /   |   /   |  \..
  \  |   /    |  /    |
    UInt     Int    Float
..   |    /   |   /      \..
  \  |   /    |  /
   UShort   Short
     |    /   |
     |   /    |
   UByte     Byte

The idea of this is that numbers can be converted lossless to a higher number type, so providing, say UInt and Float can be auto-converted to Double at runtime safely. The only place it's currently done is when collecting numbers across multiple columns and summing them, but I intend to reuse this logic in other statistical functions when needed, in parsing, or in JSON.

NumbersAggregator now converts numbers in its input to a common number type first, before aggregating, not relying on smart-casts anymore.
To avoid heavy reflection calls, types can be supplied to aggregateMixed() if you're aware of them.

[AndreiKingsley]

Looks nice! Need to test this new logic when get started with other statistical functions.

[zaleslaw]

Really intersting innovation and extensible, thanks for graph theory puzzle here!

[Jolanrensen]

I'll rename common number types -> unified number type before merging, because I think that better reflects the idea. A "common type" of Int and Double could be interpreted as using Kotlin's type system, so then it'd be Number... However, a "unified type" sounds like a new concept, which is what we do here