[SYCL] fix for leaking commands when exception thrown

[cperkinsintel]

When enqueueing a command and its dependencies, an exception might be thrown. In that case, the command will have a failed EnqueueStatus. During the clean up, we don't want to reenqueue it if we know it has failed before.

[sergey-semenov]

Could you please add a unit test for this?

[cperkinsintel]

@sergey-semenov - The mem tests ( UR_L0_LEAKS_DEBUG=1) for max_num_work_groups.cpp is how this was discovered, in its error case. I can make a new e2e test file that explicitly cribs that combination. Would that work? Or do you want a UNIT test as opposed to an E2E one? Not quite sure what that'd look like.

[sergey-semenov]

I mean, since there's no need to involve any backend to verify this, I'd prefer the more lightweight option of adding a unit test here. I think inserting a failed-to-enqueue MockCommand into the graph then checking if its destructor fired after cleaning up the buffer it depends on should be doable and check what we want here. Alternatively, you could just build a regular kernel enqueue graph with queue::submit while redefining urEnqueue... to fail and just check that we make the required ur.*Release calls.