Per-chain paralellization for rJAGS backend

[martinmodrak]

Currently, the rJAGS backend has a limitation: multiple chains for a single fit are always run sequentially. This is only relevant when one runs just a few simulations (e.g. when running 3 fits, 4 chains each on a 12 core machine one would want each core to compute a single chain, but currently, only 3 cores will be used, one for each fit).

Unlike Stan, this is not supported by rJAGS natively, so one would need to split the chains between cores within our backend code.