convert & promote_rule for Char and strings

[StefanKarpinski]

Came up over here: #25515 (comment). I think we should consider adding something like:

promote_rule(::Type{Char}, ::Type{S}) where {S<:AbstractString} = S

convert(::Type{String}, c::Char) = string(c)
convert(::Type{S}, c::Char) where {S<:AbstractString} =
    convert(S, convert(String, c))

Previously when Char <: Integer this was considerably sketchier but that's not the case anymore.

[nalimilan]

That's tempting, but I'm not sure that's right to allow converting an entry into a one-element collection like that. Shouldn't we define == between strings and chars if we do that?

Do we have other use cases than accumulate(*, ['a', 'b'])? For example, ['a', "b"] would give ["a", "b"], which may or not be desirable.

[JeffBezanson]

I think this is non-breaking?

[nalimilan]

Changes to promotion affect the type of elements inside array literals, so I think it's breaking (though not terribly so).

[JeffBezanson]

True.

I'm starting to think I'd prefer a slightly different approach to promotion in general. The only cases where I really want it are arithmetic, and mixed numeric array literals like [0, 0.5, pi]. It would be much simpler if we could say it's only for arithmetic, but AFAICT we need to have [0, 0.5] give a Float64 array or we'll have huge problems. Maybe special case array constructors where everything is a Number? I'm not even sure [[1], [1.0]] needs to give a homogeneous array result.

[StefanKarpinski]

Triage: decided against.