cholesky(BlockDiagonal(Diagonal)) is broken

[mzgubic]

In #106 we didn't think we'd need this (because why not just use Diagonal rather than BlockDiagonal(Diagonal)), but it came up in one of the nightly tests. Since we can support it relatively easily, I suggest we do

julia> s1 = BlockDiagonal([Diagonal(rand(5))])
5×5 BlockDiagonal{Float64, Diagonal{Float64, Vector{Float64}}}:
 0.373779  0.0       0.0       0.0       0.0
 0.0       0.315302  0.0       0.0       0.0
 0.0       0.0       0.266955  0.0       0.0
 0.0       0.0       0.0       0.686252  0.0
 0.0       0.0       0.0       0.0       0.0014181

julia> cholesky(s1)
ERROR: MethodError: no method matching BlockDiagonal(::Vector{Vector{Float64}})
Closest candidates are:
  BlockDiagonal(::BlockDiagonal) at ~/JuliaEnvs/GPForecasting.jl/dev/BlockDiagonals/src/blockdiagonal.jl:20
  BlockDiagonal(::Vector{V}) where {T, V<:AbstractMatrix{T}} at ~/JuliaEnvs/GPForecasting.jl/dev/BlockDiagonals/src/blockdiagonal.jl:16
Stacktrace:
 [1] cholesky(B::BlockDiagonal{Float64, Diagonal{Float64, Vector{Float64}}})
   @ BlockDiagonals ~/JuliaEnvs/GPForecasting.jl/dev/BlockDiagonals/src/linalg.jl:108
 [2] top-level scope
   @ REPL[80]:1