[LFRic] bug when profiling applied within a Loop

[arporter]

Jaffery has found that applying ProfileTrans to nodes within a loop generates incorrect code. The profile_psydatatype is imported correctly but no variable with that type is ever declared. If I alter the script to make sure the profiling is put outside the loop then the associated variable is declared correctly. I suspect this is because we currently handle everything within a loop with PSyIR and everything outside with the old gen_code method.

If it isn't too hard to fix then it would be good to get this done before 3.0 is released.