I would probably do a culvert.
What to do with the mesh is problem specific. If the bridge has a significant contraction that you want to model with the 2D solution, then you would want to have cells that are much smaller than the width of the bridge and you would probably want to set the entrance and exit loss coefficient to zero since these losses will be [mostly] captured with the 2D solution.
If this isn’t so critical and you use wide cells then the 2D will not capture the expansion and contraction losses. So in this case, I would enter the entrance and exit loss coeff as you normally would.
I would set the length to whatever it is supposed to be. If you set the length to zero you won’t have any friction losses inside of the bridge.
“there is an incidence on the water profile, that is not realistic”
I’m not sure what you mean by this.