[SYCL][Graph][CUDA] Graph buffer fill node improvement

[mfrancepillois]

Graph support in the CUDA backend for graph buffer fill nodes has been improved in UR PR oneapi-src/unified-runtime#1319

[kbenzie]

@intel/llvm-gatekeepers please merge

[bader]

Post-commit failures.
e2e-lin (Intel GEN12 Graphics with Level Zero, ["Linux", "gen12"], --param gpu-intel-gen12=True, ... / Intel GEN12 Graphics with Level Zero

Failed Tests (7):
  SYCL :: Plugin/interop-level-zero-image.cpp
  SYCL :: TaskSequence/concurrent-loops.cpp
  SYCL :: TaskSequence/in-order-async-get.cpp
  SYCL :: TaskSequence/mult-and-add.cpp
  SYCL :: TaskSequence/multi-kernel-task-function-reuse.cpp
  SYCL :: TaskSequence/producer-consumer.cpp
  SYCL :: TaskSequence/struct-array-args-and-return.cpp

e2e-lin (Intel Arc A-Series Graphics with Level Zero, ["Linux", "arc"], --param matrix-xmx8=True ... / Intel Arc A-Series Graphics with Level Zero

Failed Tests (2):
  SYCL :: Plugin/interop-level-zero-image.cpp
  SYCL :: Reduction/reduction_range_3d_rw_reducer_skip.cpp

All these failures are observed on previous commit, so they are caused by this patch.
There is one exception - reduction_range_3d_rw_reducer_skip.cpp. @mfrancepillois, could you confirm that this failure not related to your change, please?

[sommerlukas]

There is one exception - reduction_range_3d_rw_reducer_skip.cpp. @mfrancepillois, could you confirm that this failure not related to your change, please?

That test also failed on unrelated changes here: https://github.com/intel/llvm/actions/runs/8271074040/job/22630465588.