Unfortunately we're not going to generate very much juice. Assuming perfect conversion and assuming that the average person burns 100 calories per hour, we're only talking about .12 watts.
It's actually about 120 watts (a food calorie is actually a kCal), but those 100 calories are being spent through a lot more than your legs. Humans burn over 1000 calories per day just by living, so the additional amount that you're burning from walking is only a fraction of that 100 calories that get burned in the hour (assuming 2 MPH). Some of that fraction of effort is directed downward while some other fraction pushes back against the belt and moves it backwards, and then of course to your point the efficiency of capture is never 100%. The law of conservation of energy also tells us that adding a motor to capture output will add resistance to the belt, requiring you to expend more effort to walk the same distance. This is because you don't exert force against a motor in place of the ground/belt, but rather in addition to it.
I think that's a bit low. According to WA 100kcal are ~120 watt hours. The real problem is of course that us programmers hang around in chairs all day and probably need about 75kcal. Slow walking may increase that by (generous) 50kcal, of which say 75% are lost via body heat etc.
