For assembly projects the current recommendation is to use FreeCAD and A2plus module. Therefore saying there must be something wrong with App::Link, as people aren't using Assembly 3 for their production work, that is in my opinion not rationale thinking. People are being rational about it and use FreeCAD and A2plus module for now in such projects, as recommended. In addition App::Link functionality is more of a multi-feature and multi-document management piece of functionality. Therefore one use case is using it for assembly purposes, but this is still far from having an actual assembly module capability.
Over the years, from the App::Link conception, a plethora of tech-savvy users did actually get involved in discussions, as a result App::Link effort matured substantially in areas like partial loading to name one. What did lack a bit was not tech-savvy users getting involved, but tech-savvy developers getting involved. If that would happen, maybe more emphasize would be put on shape/shapeless nature of the Link feature and things like that. Topics around selection changes and working with Link feature "geometry" would likely be talked more often too. It was advised to get involved on multiple occasions, but that didn't happen. This will therefore happen after App::Link gets upstreamed and that is why in my opinion there should be no more Assembly 3 related functionality upstreamed in FreeCAD 0.19 development cycle. As tech-savvy developers will now be confronted with App::Link and some feedback will likely still need to get acknowledged.
As for "a leap of faith", that comment went in the direction, we can't expect Werner to now grasp everything and to be responsible for everything in a months time. As post merge development is still expected and that is perfectly normal. We did the same with PartDewsign NEXT and only now, with App::Link introduction, things are starting to evolve again. Like some restrictions and design decisions getting revised and things like that. If we wouldn't do PartDesign NEXT then, we should have waited until this App::Link PR, or as some suggest to wait longer? That doesn't make much sense in my opinion.
For App::Link i feel that high enough level of consensus and maturing was achieved to proceed with upstreaming. As for features such as dynamic properties. I wasn't involved in discussions and don't know how mature that functionality is. Being able to add dynamic properties to C++ features, and not only to Python based ones, is a guess a good thing.