RFC: change array and range promotion rule not to convert both containers

[JeffBezanson]

This is a re-do of #12128. This time includes Ranges and some tests.

Another example:

Before:

julia> ['a':'z', 1:2]
2-element Array{StepRange{Any,Int64},1}:
 'a':1:'z'
 1:1:2    

after:

julia> ['a':'z', 1:2]
2-element Array{StepRange{T,Int64} where T,1}:
 'a':1:'z'
 1:1:2    

julia> map(typeof, ans)
2-element Array{DataType,1}:
 StepRange{Char,Int64} 
 StepRange{Int64,Int64}

[tkelman]

a or b here? could use comments

[JeffBezanson]

@nanosoldier runbenchmarks(ALL, vs = ":master")

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan

[ararslan]

I'm inclined to think that the NullableArray regression is real, given how much promotion stuff those require.

[JeffBezanson]

I can't reproduce it locally, and there doesn't seem to be much array promotion happening there. I think it's fine.

[ararslan]

¯\_(ツ)_/¯ Okay