[SYCL][Graph] Skip prefetch/memadvise opencl tests

In intel#11474 it was noted
in the documentation that the `prefetch` and `memadvise`
handler methods were not supported as graph nodes on the
OpenCL command-buffer backend.

However, we never disabled the E2E tests for this target,
which wasn't picked up by CI, as CI doesn't use the
command-buffer extension in its OpenCL config. Correct this
oversight.

sycl/test-e2e/Graph/Explicit/memadvise.cpp

@@ -1,6 +1,9 @@
 // RUN: %{build} -o %t.out
 // RUN: %if linux && (level_zero || cuda) %{ env SYCL_PI_TRACE=2 %{run} %t.out 2>&1 FileCheck %s %} %else %{ %{run} %t.out %}
 
+// Mem advise command not supported for OpenCL
+// UNSUPPORTED: opencl
+
 // Since Mem advise is only a memory hint that doesn't
 // impact results but only performances, we verify
 // that a node is correctly added by checking PI function calls.

sycl/test-e2e/Graph/Explicit/prefetch.cpp

@@ -1,6 +1,9 @@
 // RUN: %{build} -o %t.out
 // RUN: %if linux && (level_zero || cuda) %{ env SYCL_PI_TRACE=2 %{run} %t.out 2>&1 FileCheck %s %} %else %{ %{run} %t.out %}
 
+// prefetch command not supported for OpenCL
+// UNSUPPORTED: opencl
+
 // Since Prefetch is only a memory hint that doesn't
 // impact results but only performances, we verify
 // that a node is correctly added by checking PI function calls

sycl/test-e2e/Graph/RecordReplay/memadvise.cpp

@@ -1,6 +1,9 @@
 // RUN: %{build} -o %t.out
 // RUN: %if linux && (level_zero || cuda) %{ env SYCL_PI_TRACE=2 %{run} %t.out 2>&1 FileCheck %s %} %else %{ %{run} %t.out %}
 
+// Mem advise command not supported for OpenCL
+// UNSUPPORTED: opencl
+
 // Since Mem advise is only a memory hint that doesn't
 // impact results but only performances, we verify
 // that a node is correctly added by checking PI function calls.

sycl/test-e2e/Graph/RecordReplay/prefetch.cpp

@@ -1,6 +1,9 @@
 // RUN: %{build} -o %t.out
 // RUN: %if linux && (level_zero || cuda) %{ env SYCL_PI_TRACE=2 %{run} %t.out 2>&1 FileCheck %s %} %else %{ %{run} %t.out %}
 
+// prefetch command not supported for OpenCL
+// UNSUPPORTED: opencl
+
 // Since Prefetch is only a memory hint that doesn't
 // impact results but only performances, we verify
 // that a node is correctly added by checking PI function calls