Documentation around conventions on using custom covariate ids

Currently there is little to no documentation around how to provide identifiers for custom covariates.
Real covariates seem to be able to take on any big cohort id.
One aspect seems that this can lead to collisions that may lead to hidden errors that could be prevented earlier see #168.

However, from a user perspective there is no established convention on how to define these custom identifiers or what ids may be used by existing covariates.
One solution I stumbled upon may be to use negative integer values for custom covariates, as most other covariates appear to be defined by concept id or some small positive integer. However, it is difficult for me to tell if this will work more generally as a solution.

Regardless, a runtime pre-check check to see if all custom covariate ids collide with any possible other covariate ids would be very useful.

Review the existing covariate identifier handling and investigate issue #168 to understand possible collisions and current conventions. Document which identifiers custom covariates may use, including whether negative integers are valid, and define completion as clear guidance plus an agreed approach to detecting collisions before runtime.