For ride height you could have something like a forward facing tube attached to a strut so that it is just clear of the surface at your desired ride height, and connect it to a pressure switch. You'd need to do a fair amount of smoothing on this to get an average ride height given that the pickup would be being splashed by waves, but that seems doable. One front and back gives you right height and pitch inputs.
For roll, I can see two approaches. Over short timescales, you can use a gyro to detect roll acceleration and use negative feedback to resist it. This will make the vehicle slow to roll but not stop it rolling. Over longer timescales you can use an accelerometer to measure the average direction of acceleration in a lateral plane. If the vessel is in balance (so that the net thrust vector passes through the center of mass - in other words it is upright, or turning and banked over into the turn by the correct amount to balance it) then the acceleration vector will aligned with the centerline of the vessel. If the thrust vector is offset to either side then the vessel is accelerating in roll and you probably want to try to bring it back into balance. Note that these systems will only try to resist roll - they will not completely prevent it, so it will still be possible to overcome them by steering input.
Have you thought about the sort of forces/moments and speeds that your servos will need to produce? It seems to me that these might be quite high and need big servos and a correspondingly big power supply. I assume that weight is the enemy here.