Preserve substs in opaques recorded in typeck results This means that we now prepopulate MIR with opaques with the right substs. The first commit is a hack that I think we discussed, having to do with `DefiningAnchor::Bubble` basically being equivalent to `DefiningAnchor::Error` in the new solver, so having to use `DefiningAnchor::Bind` instead, lol. r? `@lcnr`