[CMDBUF] Fix incorrect handling of shared local mem args in CUDA/HIP

[Bensuo]

• Fix handling of local mem args in CUDA/HIP for command buffer update
• Add conformance tests which check updating local memory args and work size

[Bensuo]

DPC++ PR: intel/llvm#15920

[EwanC]

Code change look fine to me, but there's some HIP command-buffer CTS tests failing in CI which might not be spurious

[EwanC]

@oneapi-src/unified-runtime-maintain This is ready for review, there is a CUDA E2E CI fail but that's not reflected in the CI for the matching DPC++ PR which is green and also seems to be in other open PRs like #2263

Failed Tests (14):
  SYCL :: Basic/kernel_info.cpp
  SYCL :: Basic/kernel_info_attr.cpp
  SYCL :: Basic/launch_queries/max_num_work_groups.cpp
  SYCL :: DeviceCodeSplit/split-per-kernel.cpp
  SYCL :: DeviceCodeSplit/split-per-source-main.cpp
  SYCL :: Graph/Explicit/spec_constants_kernel_bundle_api.cpp
  SYCL :: Graph/RecordReplay/spec_constants_kernel_bundle_api.cpp
  SYCL :: GroupAlgorithm/back_to_back_collectives.cpp
  SYCL :: GroupAlgorithm/root_group.cpp
  SYCL :: KernelAndProgram/kernel-bundle-find-run.cpp
  SYCL :: SpecConstants/2020/kernel-bundle-api.cpp
  SYCL :: SubGroup/info.cpp
  SYCL :: USM/P2P/p2p_access.cpp
  SYCL :: syclcompat/util/max_active_work_groups_per_cu.cpp