auto-round

f0572fc4 - fix: support list inputs in diffusion tuning cache

fix: support list inputs in diffusion tuning cache Single-output diffusion blocks pass tensor lists to subsequent blocks, causing the dict-only cache guard to silently skip prefetch and GPU best snapshots. Normalize these inputs inside the diffusion cache to preserve the existing block data flow. Signed-off-by: changwangss <chang1.wang@intel.com>