FillArrays much slower than MappedArrays

This is pretty strange:
```julia
julia> d_fill = Normal(2.0,5.0) ^ 20
Normal(μ = 2.0, σ = 5.0) ^ (20,)

julia> d_mapped = For(1:20) do i Normal(2.0, 5.0) end
For(#15, 1:20)

julia> x = rand(d_fill);

julia> @btime logdensity($d_fill, $x)
132.057 ns (0 allocations: 0 bytes)
-38.3707078501479

julia> @btime logdensity($d_mapped, $x)
14.346 ns (0 allocations: 0 bytes)
-38.370707850147895
```

These both call the same code:

```julia
@inline function MeasureTheory.logdensity(d::ProductMeasure, x)
@boundscheck size(d.data) == size(x) || throw(BoundsError)

s = 0.0
Δs(j) = @inbounds logdensity(d.data[j], x[j])

@inbounds @simd for j in eachindex(x)
s += Δs(j)
end
s
end
```

The only difference is is the payload:

```julia
julia> d_fill.data
20-element Fill{Normal{(:μ, :σ), Tuple{Float64, Float64}}}: entries equal to Normal(μ = 2.0, σ = 5.0)

julia> d_mapped.data
20-element mappedarray(i->Main.Normal(2.0, 5.0), ::UnitRange{Int64}) with eltype Normal{(:μ, :σ), Tuple{Float64, Float64}}:
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
Normal(μ = 2.0, σ = 5.0)
```

This is really surprising to me. If anything, I'd expect FillArrays to be a little quicker, since the value doesn't change. I tried specializing the `Δs(j)` line for FillArrays, but it doesn't seem to help.

Any idea what's going on here?