This whole series can be shelved until we need out-of-order query resolution, or more correct state promotion, or the potential performance gains are implemented and proven. The downside is we can't fix the validation layer spam from layout mismatches.
I think that fixing layout mismatches (and, in general, make the validation layer output cleaner) is quite valuable. In part because it might happen that at some point a driver actually cares about that, but also because having a clean validation output makes it easier to spot relevant validation errors. Would this MR solve that problem immediately, or is it just a prerequisite? If so, do you already have the code to fix layout mismatches?