Handling of `Number` types can be unexpected

[Jolanrensen]

Currently, numbers are inferred similar to type inference in Kotlin.

For instance:

DataFrame.readJsonStr("""[ { "a": 1 }, { "a": 1.0 } ]""")

results in a column: a: Number. This can cause issues with statistics like std, as they don't support Numbers as of now: #558
It might be easier to work with if numbers would converge in the closest common ancestor in terms of information. For instance: Int + Double + Float => Double, since Doubles can store all information a Double and Int holds.

This would need to be carefully considered, however, as it's inconsistent with Kotlin's type inference.

[Jolanrensen]

A system like #1070 should help