I remember spending a bit of time thinking on this in college. I had a feeling there had to be a counterexample somewhere; but a watered-down version may still be possible (e.g. there exist portions of any manifold that are triangulable).