A certain bending of curve can be achieved by a technique often used for linearizing of PT100 sensors. You know, these circuits, which drive a constant current through the PT100 element. There, a positive feedback from output of first gain stage to PT100 sensor is provided. This increases the signal voltage the more the higher the reading of sensor is. This results in a quasi parabolic curve.

In your case, where a more cosine looking curve needs to be fabricated you can use the positive feedback from the inverted output of first gain stage. So, that the output signal is the more limited the higher the reading is.

Positive feedback here means, that the feedback signal is fed back to the non-inverting input of first gain stage, where also the PT100 element is connected to. In your case, it's the poti resistance of float, of course, which is driven by the constant current of this linearizing scheme.

For the case of a float, which is linearily changing with height (no angle representation of poti) I could achieve a bending of curve resulting in a maximum error of -6.4% and +0% of full scale. More isn't achievable by this methode.
Another disadvantage is, that this bending goes symmetrically, not being well suited to fabricate the quasi cosine, where the difference to straight line looks more like a "drop" than a parabola.

I can draw a schematic if you are interested in. Better, you publish the actual volume-versus-angle curve of your concrete application here, then I can try to fit this curve by the above methode.

Also a simple diode limiter network could be used to fabricate the quasi cosine. But these diode limiters are prone to temperature drifts. A good remedy would be to take a CA3086, use some of the built-in pn-junctions as diodes for the diode limiting network and others to fabricate a die heating oven, to keep the temperature of die constantly at +60°C.

(Not at all assuming that this was new to you...)

Kai