You might check the tag of the topic
It’s more or less dedicated to the confined user profiles of SELinux (sysadm_u, staff_u, user_u) 
I cannot tell. On one hand, your descriptions are quite generic for such specific questions. On the other hand, that’s a decision only you can make (and that also depends on your apps’ use cases and environments): as you already found out, sysadm_u allows much more by default, but is still more active within a user account compared to unconfined_u.
By default, there should be no denials if everything is set properly. If there is a denial by default, it is worth a bug report. If denials occur after you made any change, then you might need to adjust your policy.
sysadm_u is definitely not supposed to be used “by default” (as the name indicates), but it still increases security within the account compared to unconfined_u, so it would be still an increase in security. That said, for users, sysadm_u is the only thing that I assume can be developed to the point normal users can use confinement at all and it is still a big increase of process/data isolation compared to unconfined_u → the user_u/staff_u stuff is too restrictive for Fedora-type users for several reasons.
However, when you want to make great policies and optimize them to the level of working in a confined environment, well, then you might want to test them against user_u rather than just sysadm_u. The tools we have that have standardized tests with confinement always test against user_u → if your policy works against user_u, the policy is either great (or too little restrictive:). That said (again), keep in mind that there are actions that are simply not intended for user_u (and staff_u), so depending on what your apps do or need, they might not work at all that way. That’s a reason why I think user_u will never be useful for Fedora-type users (issues start with not allowing sudo/su in user_u).
In short, I think we cannot be of much help in this very case
If you can make it work with user_u, that is the best testing environment for your policies, but you need to find out if your apps’ needs can be satisfied with user_u at all.
All that said, I much appreciate developers/maintainers who want to make their tools’ policies bulletproof
Too many see this as a burden rather than as an opportunity for opmitizing their product.