Possibly found bug

Seems that now I get term by term the same expression as the reference
for ∂τ/∂F. The errors came from use of odot instead of dot42

overall ∂P/∂F si still incorrect, however. I can think of 2 hiding
places:

1. I got bit by the right-transposed formulation of Geers again
2. The pull-back is implemented wrong