Support init keyword in maximum/minimum

[timholy]

We can supply init for general reductions, but maximum and minimum do not support this keyword. This PR allows them to do so:

julia> maximum([1,2,3])
3

julia> maximum(Int[])
ERROR: ArgumentError: reducing over an empty collection is not allowed
Stacktrace:
 [1] _empty_reduce_error() at ./reduce.jl:299
 [2] reduce_empty(::Function, ::Type{T} where T) at ./reduce.jl:309
⋮

julia> maximum(Int[]; init=-1)
-1

This was motivated by writing a blog post on invalidations (JuliaLang/www.julialang.org#794), and together with some changes to Pkg to supply the init keyword it eliminates all of the reduce_empty invalidations from loading FixedPointNumbers (which extends reduce_empty).

It's worth noting one unfortunate fact: init has different meanings for reduce(+, src) than for sum!(dest, src). In the former case, it's the value for seeding the result; in the latter, it's a boolean indicating whether to re-initialize dest to the sentinel value or to keep whatever result is already there. This PR might make the situation slightly worse by supporting more cases where the conflict becomes appreciable (maximum vs maximum!). This feels like something we should fix for Julia 2.0.

Closes #35733

[tkf]

The code LGTM but it looks like the doctest failed in CI.

It's worth noting one unfortunate fact: init has different meanings for reduce(+, src) than for sum!(dest, src).

Yes, it's also unfortunate that the meaning of init is slightly different in foldl/accumulate compared to those two families.

This feels like something we should fix for Julia 2.0.

I think we can introduce new keyword arguments (and, technically, also even deprecate init) in 1.x. But I guess I should open a new issue for discussing this.

[tkf]

👍 That's what we did in JuliaArrays/StaticArrays.jl#750

[tkf]

Can we use init as a positional argument for $_fname. I guess that would shave off some compiler work (and avoid it to get in the way of the inliner)?