Hi @Filip, welcome to the forum! 
Sounds like an interesting project! 
As a starting point, I’d recommend browsing through the other posts with the vehicle-design tag, as there’s likely some relevant information and discussion there
I expect this could be mitigated somewhat by including one or more rigid fins to help resist the undesirable rotation.
My first thought in that regard was a vertical fin just before and after the thruster to help direct the flow of water through the thruster, but I’m unsure whether that would negate or exacerbate the roll effect. It might be more useful to have fins further forward, away from the direct flow of the thruster, although that does then become harder to find space for without excessively increasing the outer diameter.
Given you’re using the different thruster types for control of independent motion axes, the type differences shouldn’t have an effect on the mixing.
From your thrust factors it looks like you’re expecting the vehicle’s centre of mass to be halfway between the thrusters (e.g. each thruster contributes equally to the rotational torque and/or sideways translation), which may be challenging in the current design given you have an enclosure at one end offset by only a thruster at the other end, and the pitch/yaw thrusters are spaced at roughly a third of the way from either side.
That may be possible to balance by adding some ballast weights, but moving that mass wastes energy if it’s not essential. If you reach a point where you can no longer redistribute mass and don’t want to add more then it’s possible to compensate in software by reducing the thrust contribution factors of the thrusters that are further away from the centre of mass, but this does reduce the total force-generating capacity of the vehicle for that motion axis.
That said, if you already plan to limit the maximum thrust of those thrusters for power supply current-capacity reasons, then such software compensation could be a net benefit, and help to reduce some design challenges.
Depending on your operating conditions, it might help to point the lights outwards, rather than inwards?
Fair enough for wanting to avoid this complexity, although integration-wise it could be quite simple (just adding a servo control that extends the arm) - the main challenges would be around the mechanical design, water-suitability of the materials, and potentially the servo depth rating (depending on your requirements).
Great to hear - I’m keen to see how this progresses over time, and it may be able to serve as a reference for others looking to make their own custom designs
By the way, please feel free to make note of any times you struggle to find information you’re after. Documentation can always be improved, and while there are areas we’re already aware of that could use some love, there are certainly also areas we just haven’t realised are missing things or in need of improvement.