Make Windows CUDA Python tests deterministic

[Copilot]

The Windows CUDA job could auto-select an unintended execution provider or fall back to CPU. Its sccache path was also scoped incorrectly, preventing cache persistence.

• CUDA validation
  • Check CUDA support before downloading test models.
  • Explicitly run the Python test driver with --eps cuda.

python test/python/test_onnxruntime_genai.py `
  --eps cuda `
  --e2e

• sccache persistence
  • Define SCCACHE_DIR at workflow scope.
  • Remove the redundant build-step assignment so restore, build, and save use the same directory.

[azure-pipelines]

Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.

[Copilot]

Pull request overview

This PR updates the Windows CUDA CI workflow to make Python testing deterministic by forcing CUDA execution-provider selection and to improve sccache persistence by scoping SCCACHE_DIR at the workflow level.

Changes:

• Define SCCACHE_DIR at workflow scope and remove the redundant per-step assignment.
• Add a CUDA EP verification step before running Python tests.
• Run the Python test driver with --eps cuda to avoid auto-selecting an unintended EP / CPU fallback.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.