[SYCL] Remove redundant creation of discarded events

[sergey-semenov]

Creation of discarded events introduces significant overhead, which is why the corresponding extension has been dropped in favor of sycl_ext_oneapi_enqueue_functions.

[Copilot]

Pull request overview

This PR removes redundant creation of “discarded” sycl::event objects (previously used when callers did not need a usable event), reducing overhead and aligning the implementation with sycl_ext_oneapi_enqueue_functions.

Changes:

• Refactor USM memory ops in queue_impl to return EventImplPtr (nullable when the caller doesn’t need an event), and wrap into sycl::event at the queue API boundary.
• Update internal submission helpers to avoid manufacturing discarded events for in-order submissions.
• Remove event_impl::create_discarded_event().

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
sycl/source/queue.cpp	Adds a small helper to wrap EventImplPtr into sycl::event for public queue memory-op APIs.
sycl/source/detail/queue_impl.hpp	Changes memory-op and helper submission APIs to return EventImplPtr instead of sycl::event.
sycl/source/detail/queue_impl.cpp	Implements nullable EventImplPtr returns and removes discarded-event creation in helper submission paths.
sycl/source/detail/event_impl.hpp	Removes the discarded-event factory function.

[Copilot]

The doxygen for memset/memcpy/mem_advise still reads as if a (non-null) event is always returned, but these functions now return EventImplPtr and can return nullptr when CallerNeedsEvent is false. Please update the "\return"/API comment to explicitly document the nullptr behavior so internal callers know they must handle it.

[Copilot]

In queue_impl::submit_impl(), the new "return CallerNeedsEvent ? EventImpl : nullptr" only applies to the main path; the earlier noLastEventPath branch returns finalizeHandlerInOrderNoEventsUnlocked(Handler) directly, which can still yield a non-null EventImplPtr even when CallerNeedsEvent=false. This can violate the CallerNeedsEvent contract and can trip the assert in submitWithHandler() (ReturnEvent==false but submit_impl returns a non-null pointer). Consider routing the noLastEventPath through the same tail return logic (or at least applying the same CallerNeedsEvent filtering there).