[UR] ASan layer fix and test update

[RossBrunton]

Two fixes for asan:

• The test has been updated to run on all adapters and also skip HIP
  and OpenCL (which don't support the required VirtualMem functions).
• GetShadowMemory has been changed to CreateShadowMemory. Previously,
  shadow memory had a static lifetime, while the Context that was used
  to construct it only had a lifetime of the AsanInterceptor object.
  It has been changed and now each AsanInterceptor creates its own
  shadow memory.

[RossBrunton]

@intel/dpcpp-sanitizers-review Can I get this looked at when you get the chance? Just some fixes for issues I was having locally with the test.

[AllanZyne]

Please don't change shadow memory related codes, we did this on purpose.

[pbalcer]

Please don't change shadow memory related codes, we did this on purpose.

But the existing code creates a global object that has a different lifetime than the loader. We should, as a general rule, avoid static lifetime on objects because it creates unnecessary complexity when it comes to cleanup and loader reinitialization. Destructor ordering is a very major problem in SYCL, and we need to be careful about not making it worse.

Can you explain why the ShadowMemory object needs to have a static lifetime?

[AllanZyne]

Please don't change shadow memory related codes, we did this on purpose.

But the existing code creates a global object that has a different lifetime than the loader. We should, as a general rule, avoid static lifetime on objects because it creates unnecessary complexity when it comes to cleanup and loader reinitialization. Destructor ordering is a very major problem in SYCL, and we need to be careful about not making it worse.

Can you explain why the ShadowMemory object needs to have a static lifetime?

Okay, I understand what you want. @RossBrunton's modification is correct.

[RossBrunton]

@intel/llvm-gatekeepers Please merge.