With such low power, the splined shaft approach could probably be just one spline - a keyway. This would certainly make building it simpler.
While a cone-shape roller could slide along a splined shaft, it still would cause excessive tire wear and have very poor efficiency. The issue isn't that the surface of the cone isn't flat on the tire, as you could tilt the whole cone. The issue is that the diameter of the cone varies as you move from the 'point' to the base. This means that the speed that the cone would try to move the bike is different, from one point to another. So, if you had one inch of cone roller in contact with the tire, the wider cross section portion of the cone would be trying to move the tire faster than the thinner portion of the cone. Which would lead to extreme tire wear.
You know, rather than jumping though engineering hoops to set up dual rollers, it might make more sense to look at applying a Comet 340 CVT to the problem, and use a single 1.5 inch roller. You would end up with, effectively, a 3/4 to 1.5 inch roller, and everything in between.
Although the efficiency would be somewhat less (95% max) than with 2 separate rollers, coupled to the engine via a timing belt, you would end up with a continuous range of torque, rather than low and hi.