Let me give it a try, I’ll try to first do it without vectors and then with vectors, though qualitatively, not quantitatively and hopefully make both clearer.
First let me make a point about linear momentum though, to help justify part of the Feynman explanation. Linear momentum is conserved. If I bounce two carts against each other on a frictionless track, I can calculate the resulting velocities using the conservation of momentum and energy. But what if I make a stationary buffer on the track and bounce the cart off that? Now momentum seems to not be conserved. I have momentum towards the buffer, and then I have momentum away from the buffer. That’s a change of twice the magnitude of the momentum in question. What gives?
Well what happens is the momentum of the Earth changes. Or more simply, the Earth-cart system has conserved momentum by various elements of the Earth (without the cart) system gaining momentum.
If I stand with my feet firmly planted on the ground and hold the spinning wheel I can, with surprising effort if the wheel is spinning fast and/or is heavy (i.e has a lot of angular momentum) change the direction of the axle. But in that situation I’m transferring some of that momentum to the Earth minus wheel system.
The swivel chair isolates me, at least in part, from the Earth system. I can’t impart momentum around the vertical axis to the Earth, because trying to do so just cause me to spin on the chair swivel. I’ve made myself one of the cars on the frictionless track.
Now initially the system of me on the chair seat and the wheel only has angular momentum around the horizontal axle of the wheel, but I can use force to twist that axle out of the horizontal. We can view that momentum as the sum of momentum around that original horizontal axle, which is now just a reference direction, a similar reference axle that is vertical.
To conserve angular momentum there needs to some element zeroing out the angular momentum around the vertical axis that I put into the wheel. That element is the whole system, me, the wheel and the chair seat, rotating in the opposite direction.
If the wheel was rotating away from me at the top, and I twist it so the top tilts to the right, there is now some small clockwise motion to the wheel when viewed from above. So I, the chair, and the wheel have to swivel anticlockwise on the chair. If I twist it the other way, the chair swivels the other way. If the rotation of the wheel was opposite, the swiveling would happen in the other direction.
Now let’s say that instead of one wheel there are two equal wheels parallel on the axis I’m holding and I spin those up in opposite directions. Now, when I twist the axle and look down from above, one wheel has gained a component around the vertical reference axis that is clockwise, but the other one has gained a component that is counter-clockwise. There is no need for the wheels, me, chair system to swivel, because those two changes cancel each other out.
And I think I’ll save explaining how vectors make this simpler for another post, as I just realized that one reason they’re almost always used is that physics curricula usually have students use them extensively before getting to this point.