Should we allow empty cubes?

[tpluscode]

Right now the very basic validation profiles require that a cube has at least one observation. It makes sense at first glance. After all, cube without data is pretty useless but that constraint has been a small hurdle for me on two occasions

1. Validating cube metadata fails when only the "head" is loaded
2. BLW voiced a requirement to be able to quickly retract invalid data whilst keeping the cube visible online

What do you think about making this only a warning instead of error if a cube happens to have no observations?

@Rdataflow @l00mi @ktk