Matrix-vector multiplication fails when DArray is unevenly distributed

The following example creates DArrays `A` and `x`, then attempts `A*x`. The only unusual bit is that `A` is distributed unevenly.

```
# test.jl

using Distributed, DistributedArrays
@everywhere using Distributed, DistributedArrays

A2 = @spawnat 2 ones((2,4)) # 2x4 array
A3 = @spawnat 3 ones((3,4)) # 3x4 array
A = DArray(reshape([A2,A3], (2,1))) # 5x4 array

x = dones((4,1)) # 4x1 array

A*x
```

Launching the REPL with `julia -p 2` including `test.jl` yields
```
julia> include("test.jl")
ERROR: LoadError: ArgumentError: cuts of the first dimension of the output matrix must match cuts of dimension 1 of the first input matrix
Stacktrace:
[1] _matmatmul!(C::DArray{Float64, 2, Matrix{Float64}}, A::DArray{Float64, 2, Matrix{Float64}}, B::DArray{Float64, 2, Matrix{Float64}}, α::Int64, β::Int64, tA::Char)
@ DistributedArrays ~/.julia/packages/DistributedArrays/fEM6l/src/linalg.jl:209
[2] mul! (repeats 2 times)
@ ~/.julia/packages/DistributedArrays/fEM6l/src/linalg.jl:262 [inlined]
[3] *(A::DArray{Float64, 2, Matrix{Float64}}, B::DArray{Float64, 2, Matrix{Float64}})
@ DistributedArrays ~/.julia/packages/DistributedArrays/fEM6l/src/linalg.jl:279
```

Looking at the source around /src/linalg.jl:279,
```
function Base.:*(A::DMatrix, B::AbstractMatrix)
T = Base.promote_op(_matmul_op, eltype(A), eltype(B))
C = DArray(I -> Array{T}(undef, map(length, I)),
(size(A, 1), size(B, 2)),
procs(A)[:,1:min(size(procs(A), 2), size(procs(B), 2))],
(size(procs(A), 1), min(size(procs(A), 2), size(procs(B), 2))))
return mul!(C, A, B)
end
```
I think the issue may be that creation of the output array `C` doesn't account for the (admittedly unusual) possibility that `A` is distributed in uneven chunks.