make I consistently provide identity behavior in type

[Sacha0]

Presently I is UniformScaling{Int64}(1), which consistently provides identity behavior in value but inconsistently in type:

julia> typeof(Float32(1)I) # as expected
UniformScaling{Float32}

julia> typeof(Float32(1) + I) # as expected
Float32

julia> typeof(fill(Float32(1), 3, 3) + I) # as expected
Array{Float32,2}

julia> typeof(fill(Float32(1), 3, 3)I) # as expected
Array{Float32,2}

julia> typeof(Int32(1)I) # expect Int32, get Int64?
UniformScaling{Int64}

julia> typeof(Int32(1) + I) # expect Int32, get Int64?
Int64

julia> typeof(fill(Int32(1), 3, 3) + I) # expect eltype Int32, get eltype Int64?
Array{Int64,2}

julia> typeof(fill(Int32(1), 3, 3)I) # expect eltype Int32, get eltype Int64?
Array{Int64,2}

julia> typeof(fill(true, 3, 3)I) # expect eltype Bool, get eltype Int64?
Array{Int64,2}

This pull request makes I yield UniformScaling{Bool}(true) instead, which consistently provides identity behavior not merely in value but also in type. Ref. #23156, and particularly the latter paragraphs in #23156 (comment), which this pull request partially addresses. Best!

[fredrikekre]

Technically breaking since typeof(I) is changing. Perhaps worth a NEWS entry?

[Sacha0]

Technically breaking since typeof(I) is changing. Perhaps worth a NEWS entry?

Good call! Rebased out conflict and added a news entry. Thanks! :)

Given the approvals and 👍s/❤️s from relevant parties, absent objections or requests for time I plan to merge this change this evening or tomorrow morning. Best!

[StefanKarpinski]

This is the only bit that gives me pause; I wonder if we should have prod([true,true,true]) === 1 instead of true. We already have sum([true,true,true]) === 3 so it would kind of make sense for the type of prod to match, even though 0 and 1 would be the only possible value.

[Sacha0]

Perhaps warrants a dedicated issue? :)

[StefanKarpinski]

Done: #24425

[Sacha0]

Thanks all!