missmatches and divB in higher resolution mhd amr runs

[UCaromel]

Most likely related to #1082 . We were seeing some mismatches (specifically in B) and divB on L1 on higher resolution runs. This accuses floating point errors as the same runs at lower resolution were not showing this kind of divB.
Similarly to #1082 where computing faraday on the ghosts what creating missmatches and divB in amr, the problem is likely caused by the upwind_constrained_transport.hpp using which uses values computed on ghosts in godunov_fluxes.hpp.
The missmatches and divB dispeared using the max_battle (https://github.com/PHAREHUB/PHARE/pull/1115/files).
However, we would like to to investigate where the missmatches appearing (seemingly when we do computation on the ghosts) really come from.