The short summary of the error (discussed on discourse and github) is that inserting PHIs with an Instruction as the insert position rather than an iterator can result in them being inserted after debug records that appear between the last PHI and the first instruction, which is a verifier error (when we convert back to intrinsics at least). There is an assertion for this, so any instances of the error should fail fast with a trivial fix (use the InsertAtHead bit, ideally by inserting at BasicBlock::getFirstNonPHIIt()). Outside of the assertion failure, consequences of using the wrong insertion method are limited to instructions being inserted after a block of debug records when they should be inserted before.
I missed a step here
I was just stating the case as I saw it for both sides (removing or deprecating). I believe that the right action would be to mark the functions as deprecated for the next release, with documentation that clearly explains the problem and how to avoid it.