add a negative test for an overlapping clEnqueueSVMMemcpy #2372

Found as part of the unified SVM discussions: We do not currently have a negative test for an overlapping clEnqueueSVMMemcpy.

We should test:

• A case where the end of the source memory partially overlaps the beginning of the destination memory.
• A case where the beginning of the source memory partially overlaps the end of the destination memory.
• A case where the source memory completely overlaps the destination memory.

Each of these cases should return CL_MEM_COPY_OVERLAP, as per the spec.