Let's see, for 104 keys 3/4" wide, the fully-rotary keyboard would have a circumference of 78" and a diameter of ~25". At that size, it should probably be oriented like a steering wheel.
The holes on my rotary phone are about 1 cm. One can make them nested, since it would have to be custom anyway: starting with the diameter of, say, 3 cm, adding a bit more than 1 cm on each side (3 cm, 4.5 cm, etc), and assuming that at least floor (pi * (1.5 + 2.5 * n - 2) / 1.5) keys fit in the nth such nested ring (1.5 + 2.5 * n is the outer diameter, - 2 is there since the holes won't be on the outer diameter, / 1.5 cm is to leave some space around them; that's not meant to be optimal, just a quick estimate), that'd yield 101 holes/keys with 6 rings, with the outer diameter of 1.5 + 2.5 * 6 = 16.5 cm for the last one. That's a rather small keyboard.
Edit: nested rings would also work for modifier keys.
I think the most straightforward way is to have a separate usual mechanism for each ring, stacked vertically, passing the rotation from each ring through the inner ones without rotating them. But likely something nicer (more optimal) can be designed: either purely mechanical (to end up with just a single combined rotation on the output) or relying on more modern technologies (plenty of options on reading dial positions then).
Or the holes can just be staggered (so that each hole is at an unique angle): then the rings won't have to rotate separately, though the reading would have to be more precise/higher-frequency then, and it may be tricky to rotate the dial with a finger as precisely (with 6 rings and 15 mm per hole, that would be a 15 / 6 = 2.5 mm margin, roughly).
It doesn’t make sense for the modifier keys to be on the rotary dial, but the rest you code encode by a pair of dials (for 10x10 keys). Labeling would be a challenge.
I wonder which of Emacs and Vim would have the advantage here.
https://site.xavier.edu/polt/typewriters/simplex.html