Alina,
I am not going to be able to help much. I have not had the opportunity to try the new z-enabled capability of Collector for ArcGIS (v19.0.1). This is primarily supported with the Arrow/EOS receivers and software and we are using Leica receivers that, unfortunately, have not caught up yet. Your workflow outline appears to be sound based upon my understanding.
The main issue with data obtained from Collector is "What coordinate system is my data in?" Going from a GPS receiver, running on various RTK systems, on different coordinate systems, that each manufacture integrates differently with Collector, pushing to AGOL, that uses base maps in one coordinate system but allows you to publish your features in another, back to ArcGIS Pro that does not clearly define which coordinate system it is currently working in, and the fact that there are hundreds of possible combinations of coordinate transformations, all to have your data not appear where you think it should be appearing, or is the data right and the map/image wrong. No wonder it is confusing.
One thing I can highly recommend is to create a set of control points. I have two sets I use, one is a small number of points around the office neighborhood that I can quickly go for a walk, on a nice spring day, and test a new collector map. The second is about 24 points over a large area that I can use to test at finer level that can catch things such as datum shifts that may not be visible on a small localized set of points. We collected these control points using a survey grade GPS, using the coordinate system we ultimately will be working with and typically using 5-10 minute averaging at each point. The points have been collected at several different times and the results averaged. I keep them in a shapefile or GDB feature class that is readily accessible.
When I am working with a new Collector Map or trying to figure out is my new transformation workflow model working correctly, I go for a walk and collect these points. Once collected I bring only the points into an ArcGIS session, perform my transformation workflow and then bring in my controls points to see how they compare. Sometimes my collected data shows up in the middle of the Indian Ocean, other times it is a few feet off but most of the time it is within the accuracy of the GPS. When it does not work I tweak my workflow and run it again, check my map or GPS configuration, recollect the points and check it again. It is simple and quick and I am checking against a single absolute known.
Working with multiple coordinate systems in you data is difficult. We are in the process of migrating from NAD 83(86) to NAD 83(2011) the difference between the two systems can be 2+ feet. The key here is to make sure you have the correct transformation assigned in ArcGIS between the two systems. There is also a performance price to pay for working with multiple coordinate systems as some of your data needs to be projected on the fly. This is most notable when it is your imagery that is being corrected. The best option is to work in a single system or as a second option work in your imagery coordinate system to minimize processing time. Another option, if possible is to obtain a physical copy of the imagery, is to reproject the imagery to your preferred datum. Just plan on some serious processing time depending upon the extent and image quality.
Sorry I cannot specifically answer your question. I hope this helps.