svm_set_kernel_exec_info_svm_ptrs may access an SVM allocation incorrectly

I believe that the test for `svm_set_kernel_exec_info_svm_ptrs` is incorrect in at least some cases, specifically how it acceses the `pBuf` SVM allocation for the call to `clSetKernelExecInfo`:

https://github.com/KhronosGroup/OpenCL-CTS/blob/9fc0d23b4cfccd84be8927363a77107dc554de30/test_conformance/SVM/test_set_kernel_exec_info_svm_ptrs.cpp#L107-L115

Note:

* `pBuf` is a (coarse-grain) SVM allocation. This means that it must be mapped and unmapped to access the data on the host.
* `pBuf` itself is never mapped or unmapped, but there is a `bBuf` OpenCL buffer object that is created with the `pBuf` coarse-grain SVM allocation, which is mapped and unmapped to populate the SVM allocation.
* The command to unmap the `bBuf` OpenCL buffer object has been enqueued at the time of the call to `clSetKernelExecInfo`, so the buffer object may be unmapped (and therefore, `pBuf` may be inaccessible) at the call to `clSetKernelExecInfo`, depending whether the unmap command is executed greedily or lazily.

Here are two possible ways to make this test safe in all cases. There may be others, too:

1. Call `clSetKernelExecInfo` before unmapping `bBuf`, where the `pBuf` pointer is guaranteed to be accessible.
2. Pass a separate host pointer to `clSetKernelExecInfo` rather than reusing `pBuf` for this purpose.

Inspect test_conformance/SVM/test_set_kernel_exec_info_svm_ptrs.cpp around lines 107-115, focusing on how pBuf and bBuf are used before clSetKernelExecInfo. Verify the test does not access the coarse-grain SVM allocation while it is unmapped, then run the relevant SVM test to confirm the corrected ordering or pointer usage.

Written by the indexing model from the issue text.