Fix two small things in FiniteElement.

[drwells]

Taken from #14583.

1. Avoid hard-coding in the number of permutations
2. Correctly calculate the combined/raw orientation

Here and elsewhere, we store the orientation information in a single unsigned char which we unpack to call this function and then pack again to access the array. Is there any reason to not just pass the combined orientation as an argument? I'd like to deprecate this function and eliminate that middle step to avoid packing/unpacking bugs like this one.

[drwells]

This fix might not be right. Pre-simplices we had

  return index+adjust_quad_dof_index_for_face_orientation_table(index,4*face_orientation+2*face_flip+face_rotation);

which is not the right packing (and won't work for simplices since those orientations go 0-5, so these weights on orientation + flip are out of bounds) but is perhaps consistent with other parts of the library. Lets see what the CI does.

[bangerth]

This is apparently not the same encoding used in other places in this class -- judging by the fact that a number of tests fail. You're completely correct that we ought to not do the en/decoding of these flags by hand, but using the corresponding functions in ReferenceCell. Maybe the right step would be to just search through the entire file for face_flip or some such, and replace all of these hand encodings by what ReferenceCell offers.

[drwells]

Definitely - this would be a good time to simultaneously make our encodings uniform and do the follow-ups listed in #12878. We have at least three different orientation conventions in the library and I would much rather have one.

[bangerth]

Yes, three is too many when there are really only 6 possible (reasonable) encodings of three bits of information...

[drwells]

These orientation fixes are going to be quite large - I'll handle them in separate future PRs.