[CI] Fix cuda_dev_kit detection on runners (#16892)

The CUDA dev kit is already installed in the image we use but we don't
pass these CMake vars so it isn't found.

It should be fine to do this unconditionally, lit.py does a full compile
with required files and checks the compiler return code, so even if it
isn't actually installed it won't cause any problems. The compile is
failing today because we don't pass the vars anyway.

Manually tested this on the runners.
Confirming working in
[this](https://github.com/intel/llvm/actions/runs/13164735078/job/36742725212?pr=16892)
run, as `Adapters/cuda_queue_priority.cpp` passed which requires the
feature.

Signed-off-by: Sarnie, Nick <[email protected]>

Author: sarnex

devops/actions/run-tests/e2e/action.yml

@@ -55,7 +55,7 @@ runs:
     if: inputs.testing_mode != 'run-only'
     shell: bash
     run: |
-      cmake -GNinja -B./build-e2e -S./llvm/sycl/test-e2e -DCMAKE_CXX_COMPILER="${{ inputs.cxx_compiler || '$(which clang++)'}}" -DLLVM_LIT="$PWD/llvm/llvm/utils/lit/lit.py" ${{ steps.cmake_opts.outputs.opts }}
+      cmake -GNinja -B./build-e2e -S./llvm/sycl/test-e2e -DCMAKE_CXX_COMPILER="${{ inputs.cxx_compiler || '$(which clang++)'}}" -DLLVM_LIT="$PWD/llvm/llvm/utils/lit/lit.py" -DCUDA_LIBS_DIR=/usr/local/cuda/lib64 -DCUDA_INCLUDE=/usr/local/cuda/include ${{ steps.cmake_opts.outputs.opts }}
   - name: SYCL End-to-end tests
     shell: bash {0}
     env: