add tests for differentiability

[baggepinnen]

This PR adds tests to make sure all Kalman filters are differentiable w.r.t.

• R1
• R2
• p in the dynamics

The important thing to ensure that filters are differentiable is to let the initial state distribution d0 have the correct element type, e.g., something like

function costfun1(p::AbstractArray{T}) where T
    d0 = MvNormal(@SVector(zeros(T, nx)), T(1.0))   # Initial state Distribution with correct element type
    # construct filter
    ...
end