Support keyword arguments in mean

[BatyLeo]

Support for kwargs was missing in the mean method with a preprocessing function f.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Files	Coverage Δ
src/distribution.jl	100.00% <100.00%> (ø)

[gdalle]

I didn't add it because the Statistics version doesn't support kwargs, so it felt counter-intuitive that ours would? Users can always fix kwargs themselves.

julia> using Statistics

julia> f(x; y) = x + y
f (generic function with 1 method)

julia> mean(f, 1:3; y=2)
ERROR: MethodError: no method matching mean(::typeof(f), ::UnitRange{Int64}; y::Int64)

Closest candidates are:
  mean(::Any, ::AbstractArray; dims) got unsupported keyword argument "y"
   @ Statistics ~/.julia/juliaup/julia-1.10.4+0.x64.apple.darwin14/share/julia/stdlib/v1.10/Statistics/src/Statistics.jl:104
  mean(::Any, ::Any) got unsupported keyword argument "y"
   @ Statistics ~/.julia/juliaup/julia-1.10.4+0.x64.apple.darwin14/share/julia/stdlib/v1.10/Statistics/src/Statistics.jl:61
  mean(::Any) got unsupported keyword argument "y"
   @ Statistics ~/.julia/juliaup/julia-1.10.4+0.x64.apple.darwin14/share/julia/stdlib/v1.10/Statistics/src/Statistics.jl:44

[BatyLeo]

I added this because I needed it for InferOpt.
By fixing kwargs themselves, do you mean something like this:

f(x; y) = x + y

mean(x -> f(x; y=2), 1:3)

[BatyLeo]

Alternatively users can use FixKwargs, I agree it's not needed then.

[gdalle]

DifferentiableExpectation can expose FixKwargs as part of the public (unexported) API if it helps

[BatyLeo]

I think it's a good idea, I'm already using it extensively in the current InferOpt overhaul