Improve combination tests

[SamuelBrand1]

We are still adding unit tests (see #179 and #180) but are lacking tests that cover combinations of model components (e.g. combinations of latent models and observations models) and we are lacking tests that cover multiple AD paradigms (e.g. ForwardDiff, tape-based reverse AD, source-based reverse AD etc).

[seabbs]

I think you have some work on this @SamuelBrand1 is it in a WIP branch?

[SamuelBrand1]

We've resolved:

• AD tests (covered by TuringBenchmarks).
• We now test a combination set of IGPs and latent models whilst fixing a delay observation model.

Q: Does #371 close this and we have new issues for the newer models going forwards?

[seabbs]

yes I tihnk so