Internally any vertex on a hard edge will be split into multiple vertices on a hard edge. So for example, a corner of a cube is not a single vertex with three normals, it is actually three vertices with one normal each. Also remember that vertex shaders operate on a single vertix at a time, so you can’t simply average these normals across the different split vertices.
If there is a specific effect you want to achieve that requires this you could potentially do some kind of preprocessing on the mesh and bake a second set of normals into the vertex color channel. However keep in mind that the vertices are still split and you are adding additional vertex color data, so your mesh would be quite a bit heavier. So if your goal is to simplify your pipeline and you do not want to be able to blend between the two I would definitely recommend modifying the source mesh, it’s not like that is a lot of work and if you happen to have to process a large number of assets, then writing a script to batch process them is probably less work then getting this to work.