drake visualizer has normals flipped for box primitives

[RussTedrake]

our current manipulation_station example uses a lot of box primitives for visualizing the station, etc. and I've been noticing that the normals on the box primitive, as rendered through the drake_visualizer interface are consistently bad. so far, it's only made for aesthetic problems, but now I'm trying to manipulate boxes and the objects often dissappear into the table / surrounding structure.

This is the station viewed from a nominal (human-relevant) viewing angle. It's supposed to be a table with 80-2 vertical trusses. But the closest leg is disappearing into the base (+ other artifacts)

this is the same station, viewed from a slightly lower camera angle (but still above the table/ground). now you can see that there was actually a "foam brick" on the table the entire time. it was just being hidden by the visualization artifacts.

you can also see into the sides of the table surface, which should be a solid box. is there a reason that this might have changed? vtk version bump?

[EricCousineau-TRI]

Running //examples/manipulation_station:proof_of_life on my machine on current master (81352d8) did not indicate odd normals:

Perhaps this is specific to how the box primitives are making it in? Do you have a reproduction model + script that it's easy to run (perhaps a pydrake script using ConnectDrakeVisualzier + an SDF file)?

Also, is the alpha value set to 1.0? (I believe Z order drawing gets weird when it's not purely opaque...)

[RussTedrake]

Nice call! The alpha values were 0.9. Setting them to 1.0 has resolved it. You rock.

cc @pangtao22 , @Sproulx -- let's be sure to use alpha = 1.0 from now on.