[Java] Reconsider dedicated setting wrappers (setNumGpus, setMpdlpPartitioner, ...) vs generic setSetting

[ramakrishnap-nv]

SolverSettings currently has both a generic setSetting(String name, ...) API and a growing set of dedicated wrappers (setMethod, setPDLPSolverMode, setOptimalityTolerance, setNumGpus, setMpdlpPartitioner). Each new parameter that gets a named wrapper adds ongoing API surface and maintenance burden.

Worth deciding whether new settings like num_gpus/multigpu_pdlp_partitioner should stick to the generic setSetting(CuOptConstants.CUOPT_NUM_GPUS, value) form rather than growing a dedicated method per parameter, and possibly trimming existing wrappers to match.

Raised during review of #1984.

[ramakrishnap-nv]

Opened ramakrishnap-nv#5, stacked on #1984 since it touches the same lines. Will retarget to main once #1984 merges.

[ramakrishnap-nv]

Opened #2000 (draft, stacked on #1984 for now).