`KeyedDistribution` allows mismatch between the keys and the underlying distribution keys

```julia
julia> ka = KeyedArray(rand(3), down=["a", "b", "c"]);

julia> mvn = MvNormal(ka, [1, 1, 1]);

julia> mean(mvn)
1-dimensional KeyedArray(NamedDimsArray(...)) with keys:
↓ down ∈ 3-element Vector{String}
And data, 3-element Vector{Float64}:
("a") 0.8185032515167808
("b") 0.2940895403871865
("c") 0.7714754947104043

julia> kd = KeyedDistribution(mvn, [1, 2, 3]);

julia> mean(kd)
1-dimensional KeyedArray(NamedDimsArray(...)) with keys:
↓ _ ∈ 3-element Vector{Int64}
And data, 3-element Vector{Float64}:
(1) 0.8185032515167808
(2) 0.2940895403871865
(3) 0.7714754947104043
```

Should this be allowed? It bit me recently where I've accidentally constructed the KeyedDistribution without naming the dimension. Maybe we want to check this when constructing the KeyedDistribution?

(Similar thing for `var`)

Similar to https://github.com/invenia/KeyedDistributions.jl/issues/17