weighed mean/sum fail on vector of vectors

[oschulz]

StatsBase currently fails to calculate weighted sums and means of vectors of vectors/arrays (unweighted sums/mean works fine).

Given

A = [[1.5, 2.5], [3.5, 4.5]]
w = Weights([0.5, 0.6])

The unweighted mean does what it should:

julia> mean(A)
2-element Array{Float64,1}:
 2.5
 3.5

However, the weighted mean fails:

julia> mean(A, w)
ERROR: DimensionMismatch("x and y are of different lengths!")

This happens because StatsBase defines mean(A, w) via sum(A, w) / sum(w), and sum(A, w) via dot(A, values(w)). However, dot operates recursively, and so tries to do dot(A[i], values(w)[i]) with each element of A (elements are vectors) and w (elements are scalars).

I believe this is a valid use case - @lmh91 and I encountered this while trying to calculate the weighted mean of 3D spatial positions (represented as a vector of static vectors).

[oschulz]

Dup of #518.