Reading down your comprehensive list above, I am struck by why I never seem to encounter these issues in hours and hours of my own unwrapping, and I think we are using different methods !
The 'rules' I follow are thusly:
1. Only do automatic unwrap on sections of mesh clearly made for it.
2. Define poly selections, and project those frontally from all the best angles, then use 4, 5 and 6 keys to put those and align them where I want them.
3. Optionally Pin and Relax those meshes, optionally using cuts I have saved to a selection tag
4. Activate UV options to show Distortion (100%) and Island Connectivity (now I can see what connects where)
5. Use 'weld and relax' mainly in poly mode (instead of weld, on the whole) or manual UV point terracing to combine islands if necessary (and I tend to move the islands adjacent to where I want them to end up before doing this).
6. Further relaxment if I went the manual terrace route.
7. UV Packing.
You will detect something of the 'legacy' in my workflow, and that is true - I liked the 'old' way of unwrapping stuff, and that continues to work for me very nicely in Cinema, enhanced as it is by the new layout and UV mode details. Occasionally I find that approach isn't optimal or takes a bit long for really organic stuff, so only then do I head over to the newer cut and unwrap tools like the Ministry of flat-based UV Unwrap etc. And because I don't use that much I guess that may be why your specific issues are new to me ! 🙂
You are right that in my example above the red areas are less than ideal (and that is the best Cinema's relax routines can manage using workflow above), but the Magnet (UV) and brush relax (modelling) tools are there to sort out situations like that, where it is important using the old workflow.
If you have connectivity turned on you usually can see what will weld where. The UV Weld tool does not move islands first, and I believe that is intended. If you want to move islands at the same time, use weld and relax instead. The help is very illuminating about precisely how that works, but in trying it myself I have found a few bugs and moments of unexpected behaviour...
I can't answer your questions on texel density or UDIMs alas (though fairly sure Maxon is aware of and working on those) - as a mainly modeller I have so far considered those 'none of my business' , and there are people much better qualified than me to comment on those aspects.
CBR