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