UR Offload: align enqueue calls with updated OffloadAPI signatures

[Copilot]

The Offload job in UR precommit regressed due to API drift between the offload adapter and the current OffloadAPI.h function signatures. The adapter still used old call forms for event creation and kernel launch, causing compile-time argument mismatch errors.

• What changed

  • olCreateEvent call sites updated: pass explicit event flags at all enqueue/event creation paths in enqueue.cpp.
  • olLaunchKernel call updated: pass the new launch-properties argument (nullptr for current behavior) to match the extended signature.
  • Default event flags centralized: introduced a local constant to avoid repeated literals and keep call sites consistent.

• Why this resolves the CI failure

  • Build break source removed: all offload enqueue paths now match current Offload API prototypes, eliminating the type/arity mismatches reported by the failing job.

constexpr ol_event_flags_t OL_DEFAULT_EVENT_FLAGS = 0;

olCreateEvent(Queue, OL_DEFAULT_EVENT_FLAGS, &Event->OffloadEvent);

olLaunchKernel(Queue, hQueue->OffloadDevice, hKernel->OffloadKernel,
               hKernel->Args.getStorage(), hKernel->Args.getStorageSize(),
               &LaunchArgs, nullptr);

[ldorau]

CI failures:
https://github.com/intel/llvm/actions/runs/26454604862/job/77887612963?pr=22115
https://github.com/intel/llvm/actions/runs/26454634457/job/78036909296?pr=22092

[Copilot]

Pull request overview

This PR fixes ur_adapter_offload link-time failures when consuming a prebuilt LLVM Offload install whose libraries are placed under multiarch lib directories (e.g., lib/x86_64-unknown-linux-gnu) instead of a flat ${prefix}/lib layout.

Changes:

• Replaces a hardcoded ${UR_OFFLOAD_INSTALL_DIR}/lib/libLLVMOffload.so link with a resolved UR_OFFLOAD_LLVMOFFLOAD_LIBRARY path.
• Adds CMake logic to locate libLLVMOffload via find_library() with common libdir suffixes, plus a glob fallback for nested multiarch layouts.
• Emits a fatal configure error if libLLVMOffload cannot be found.

[ldorau]

@lplewa please review if this fix makes sense (or if we can close this PR)