Stick to surface will be probably implemented using VTK picker classes, which can find 3D position of a point visible in a view at a chosen position. These picker classes already work with most actor types, such as polydata models and volumes. There is no limitation here.
You are right that VTK locator classes require mesh inputs. However, locators are not needed (and not well suited) for 3D position computations for “stick to surface” feature. If for some reason you needed locators for 3D volumes anyway, then you could generate the necessary iso-surface very quickly.