Unify `par_dispatch`, `par_for_outer` & `par_for_inner` overloads #1281

Job	Run time
perf-and-regression (cuda, serial)	27m 1s
perf-and-regression-amdgpu (serial)	14m 37s
perf-and-regression (cuda, mpi)	1h 15m 37s
perf-and-regression-amdgpu (mpi)	18m 38s
perf-and-regression (host, serial)	1h 13m 33s
perf-and-regression (host, mpi)	32m 7s
	4h 1m 33s

Provide feedback