1. Correct - all 3 RGs will fail if Untrust is down, as long as it is tracked with a weight of 255 in all 3 RGs.
2. Correct - with preempt enabled, as long as there is an active failure (or failures) with a failure weight of 255 the RG will remain on the secondary. Once the failure condition is cleared (in this case, Untrust comes back up), then the RGs will all fail back to the primary.
3. This gets more into a question of style and preference than an absolute technical recommendation. I've heard people say yes and I've heard no. Even with preempt I prefer to set up monitoring on both chassis - it doesn't really have any effect if you have preempt enabled, but that way if I disable preempt at some point in the future then I don't have to remember that I didn't set up full bidirectional failovers.
Note for #2 above that RG0 (the control plane) will fail independantly of the forwarding planes (RG1+) and does not support preempt; it can continue to run on the secondary unit with typically no issue. I say 'typically' because there are some instances where it matters. UTM/IPS in pre-11.4 code for Branch would only run on the chassis that had the active control plane; on the high-end, LSYS only runs on the chassis with the active control plane. Those are the only examples I can think of where you need to keep the control plane on the same chassis as the forwarding planes, and if you're not runnign them then you can ignore this. To keep RG0 together with RG1+, I typically disable preempt for the forwarding planes (RG1+) and enable interface tracking for the control plane (RG0) on both the primary and secondary units. This would have all RGs failing from one chassis to the other in sync; you would remain in active/passive (you had some scenarios where active/active would be in effect) but UTM (pre-11.4) or LSYS would work for all RGs.