Faster multigrid solve using shared memory

[AlexanderSinn]

This PR combines four Gauss–Seidel iterations into a single GPU kernel using shared memory.

Test with 4095*4095*1000 cells, time taken for HPMG:

A100:
old: 32.64 s
new: 14.91 s (+120% speedup)

MI250X:
old: 49.35 s
new: 19.87 s (+148% speedup)

• Small enough (< few 100s of lines), otherwise it should probably be split into smaller PRs
• Tested (describe the tests in the PR description)
• Runs on GPU (basic: the code compiles and run well with the new module)
• Contains an automated test (checksum and/or comparison with theory)
• Documented: all elements (classes and their members, functions, namespaces, etc.) are documented
• Constified (All that can be const is const)
• Code is clean (no unwanted comments, )
• Style and code conventions are respected at the bottom of https://github.com/Hi-PACE/hipace
• Proper label and GitHub project, if applicable

[MaxThevenet]

Great speedup, thanks!