SYCL listed forces: optimize parameter passing

Compared to 2023.0, using ADH cubic. Speed-up of "Bonded" kernels:

	F-only	FV
hipSYCL 0.9.4, ROCm 5.2.5, bundled Clang, gfx90a	2.67	2.33
hipSYCL 0.9.4, ROCm 5.4.1, Clang 15.0.7, gfx1034	1.94	1.56
hipSYCL 0.9.4, CUDA 11.8, Clang 15.0.7, sm_86	1.00	1.00
IntelLLVM nightly 2023-02-06, CUDA 11.8, sm_86	4.09	1.55
oneAPI 2023.0, Intel Arc 770	1.29	1.01
oneAPI 2023.0, Ponte Vecchio	1.10	~1

Based on AMD/StreamHPC optimization.

Refs #3928 (closed), #4593 (closed)