Explore why d is being passed via theta to primary_censored_integrand

[seabbs]

On the face of it I think theta should be for estimated parameters which d is not (its data) in all cases this may be having a performance impact

[seabbs]

One of the reasons is that anything passed to x_r has be data and that this means that d has to be directly passed as data to the external stan wrapper function. This is currently a problem as these functions mutate d to add window to it.

One potential solution we could consider would be to use delay_lower and delay_upper in place of d and swindow. This would break some of the connection with the R code but should work (though it is hard to know if it is worth doing without trying).

One place this might be an issue is in the vectorised pmf function as this takes a single int and loops over the cdf. However, it may be the case that for loop index variables are privileged and so this won't be an issue.

[seabbs]

This was closed by #44 by moving d to a data arg