`mean(skipmissing(x))` is imprecise

[tiemvanderdeure]

The generic fallback for mean can give wrong results in some cases. Maybe it would be a good idea to add a specialized implementation when it is called on a SkipMissing, as there is for sum. The easiest way to do this would be to pass a function that computes both sum and count to mapreduce.

A = rand(Float32, 1_000_000) .+ 1f6
mean(skipmissing(A)) ≈ mean(A) # false
mean(A) ≈ (sum(skipmissing(A)) / length(A)) # true

Related: #30421

[mikmoore]

Heavy overlap with #52365, if not a duplicate. #52397 would fix this.

[mbauman]

Duplicate of #52365