I haven't tested this, but it should be possible in principle. When the mode expansion monitor MEM is not angled this leads to imprecision in the calculated modes and mode expansion calculations. For examnple the corss section is different and the phase of the fields will be slightly off.
At least for the mode source you can easily angle it to be normal to the waveguide, which you should do if it is relevant to your set-up.
The real challenge I think comes with setting up your FOM monitors. By default you simply specify a frequency monitor. These cannot be angled and must align with the principle FDTD planes. All rotation calculations will be done in the mode source or MEM. Since you do not define an angle in the lumopt set-up script it is undetermined in the algorithm, which adds the MEM and adjoint sources automatically.
If the FOM ports are on angled waveguides you may have some numerical issues, but it may be worth experimenting with anyway. Even though it isn't exact the trend should be consistent which is enough for optimization. That being said these gradient based optimization methods can be quite sensitrive to numerical issues, so I cannot guarntee anything. To do this rigorously you would need to add a rotation argument to the modematch python class in lumopt, which rotates the MEM correctly. All that python code is editable, so you are free to make updates if you like, but we can't really support such efforts.
Alternatively you could add this as a feature request through.