To solve our "initial value problem" I would like to add a function that adaptively creates an initial value function based on the brmsformula and the stanched generated by a bmm call.

I propose to add a property init_ranges to all bmmodel objects that would benefit from setting initial values for starting the sampling process and access this information to set initial values accordingly.

The biggest challenges I anticipate are:

1. a complex bmmformula that suppresses the intercept and includes a combination of effects: e.g. parameter ~ 0 + group + group:condition + (0 + group:condition | gr(ID, by = group))
2. non-linear formulas in the bmmformula that require to assign which newly generated parameters are linked to which model parameter. Although this issue might be more complex as non-linear formula could also include multiplicative combinations of two non-linear parameters for which initial values can not be reasonably set as a default

With respect to 2., we might consider throwing a warning that users have to specify initial values by themselves for such non-linear formulas.