[SYCL][NFC] Avoid XPASS-es in some graph tests (#12421)

A couple of graph related tests were marked as `XFAIL`ed on OpenCL
backend and it works just fine for intel/llvm. At intel/llvm we launch
E2E tests for multiple targets in one LIT run, but there are some side
effects of that mechanism like treating `XFAIL` as `UNSUPPORTED`.

We have downstream testing environments where E2E tests are launched for
a single target. When those tests are launched on an OpenCL device
without graphs support, they do early exit with exit code zero which is
equivalent to `PASS` from LIT point of view. That turns into an `XPASS`
due to `XFAIL` in the test, causing our CI to report failures.

Changed `XFAIL` to `UNSUPPORTED`, which will work for all possible
environments, regardless of amount of targets for E2E tests and status
of graphs support on OpenCL devices used in testing.

Author: AlexeySachkov

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

@@ -10,7 +10,8 @@
 // should. So the Sycl graph support cannot correctly catch the error and throw
 // the approriate exception for negative test. An issue has been reported
 // https://github.com/bashbaug/SimpleOpenCLSamples/issues/95
-// XFAIL: cuda, opencl
+// XFAIL: cuda
+// UNSUPPORTED: opencl
 // Note: failing negative test with HIP in the original test
 // TODO: disable hip when HIP backend will be supported by Graph
 

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

@@ -10,7 +10,8 @@
 // should. So the Sycl graph support cannot correctly catch the error and throw
 // the approriate exception for negative test. An issue has been reported
 // https://github.com/bashbaug/SimpleOpenCLSamples/issues/95
-// XFAIL: cuda, opencl
+// XFAIL: cuda
+// UNSUPPORTED: opencl
 // Note: failing negative test with HIP in the original test
 // TODO: disable hip when HIP backend will be supported by Graph