[new-model] Implement full ddm

[GidonFrischkorn]

Based on the likelihood function available in STAN (v. 2.35) I would like to write a wrapper to make the specification of the full DDM a little easier. Likely this model will not run very fast, but this way we have an implementation of a DDM with support of all trial-to-trial variability parameters in the package. And the coding work is minimal.

This is thought to extent the other evidence accumulation models I am currently working on (see #277 and #281)

I like the idea of having different implementations that provide fast algorithms to fit simplified accumulations models (e.g. ezdm or cswald) but also have the functionality to estimate the full DDM if necessary.

Steps for implementation:

• fill in model_info
• implement STAN functions
• specify check_data.ddm
• specify check_model.ddm
• specify bmf2bf.ddm if necessary
• specify check_formula.ddm
• specify configure_model.ddm
• add function to generate initial values to allow for proper start of sampling
• specify distribution functions for the ddm model
• check that pp_check and bridgesampling works for cswald
• if possible: optimize sampling and speed for model estimation

[GidonFrischkorn]

I have added the ddm in some commits to the associated branch. And the generation of the initial values seems to work well with the create_initfun method I have added on #290

[GidonFrischkorn]

While sampling starts fine with cmdstanr as a backend, sampling with rstan throws an error because of a dimensions mismatch of the expected and passed initial values. For now, I do not really understand where the problem is, as the stancode generated for the models and thus also the parameters are the same irrespective of the backend.

So, I suspect that rstan somehow treats the initial values differently.

[GidonFrischkorn]

I just tested and the problem is model independent. That is for the sdm initial values also only work with cmdstanr as backend, but rstan declares a similar dimensions mismatch as for the ddm.