Projection transfer randomness

Bug Description

We (@muelcm and I on separate inputs) have observed that steady state calculations restarted into transients exhibit non deterministic behaviors in the projection transfers. The integrals of power density (transferred quantity) change between two consecutive runs At least in my inputs, I have made sure the variable types do not change in between the simulations so it's not this expected mode of failure.

Steps to Reproduce

Use fixed mesh on an exodus restart iirc might still trigger without fixed_mesh

Impact

Affects simulation results. Since the norm of the field is different, it's not that hard to realize there is an issue.

idaholab / moose