The model region can be on a rotated pole as far as I know, it’s just the ancillaries used to convert from GRIB to UM formats that need to be on a standard pole. Once the boundary conditions are in UM format the reconfiguration can do the transfer from standard to rotated pole. That’s what the UKV domain does, driving data is standard pole and the inner region is rotated.
You can run a global GA driving model as part of the nesting suite- it’s one of the options instead of driving from ERA5. You would need initial conditions for the date of interest. Possibly the CSIRO folks can comment on switching between CABLE and JULES variables?
I guess for free running you get more opportunity for the model to completely spin up as the state is not being reset all the time, but also the nest is more likely to drift away from the driving model without the resets. Probably the more science focussed people will be able to comment here.