Fix bug in event.cpp where queue_impl outlives async handler and causes a segmentation fault

[lbushi25]

This PR fixes a flaky bug in event.cpp where an async exception handler associated to a queue had already gone out of scope by the time the queue_impl destructor was called. Since the latter(destructor) calls the former(exception handler), this is clearly a problem as we get undefined behavior. This PR moves the exception handler, previously declared in function scope, to global scope so that it always outlives the queue resources.

[CLAassistant]

All committers have signed the CLA.

[TApplencourt]

Haha, related too the recent KhronosGroup/SYCL-Docs#925
We will discuss in the next call, just to be sure everybody have the same understanding

[TApplencourt]

In theory better to clear(); at the beginning of the test, not at the end.

Indeed if a test fail, maybe clear() will not be triggered and leave the handle in a bad state. But 🤷🏽 , not that critical.
Thanks for the fix!

[bader]

In theory better to clear(); at the beginning of the test, not at the end.

Indeed if a test fail, maybe clear() will not be triggered and leave the handle in a bad state. But 🤷🏽 , not that critical. Thanks for the fix!

@lbushi25, let me know if you are going to apply this comment.

@TApplencourt, once @lbushi25 confirms that PR is ready, I can merge it, right?

[lbushi25]

In theory better to clear(); at the beginning of the test, not at the end.
Indeed if a test fail, maybe clear() will not be triggered and leave the handle in a bad state. But 🤷🏽 , not that critical. Thanks for the fix!

@lbushi25, let me know if you are going to apply this comment.

@TApplencourt, once @lbushi25 confirms that PR is ready, I can merge it, right?

I made the changes. As for me, once I check that the CI passes, we can merge it.

[lbushi25]

@bader Feel free to merge if @TApplencourt agrees!