First commit adds coverage to a similar case to #13370 . h/t dergoegge for detecting this issue independently. Documentation of this behavior with a test is then followed by a proposed change where anytime we considered something VALID_TRANSACTIONS
and potentially advertised via compact blocks, we should respond to requests for the getdata
if the peer fell back for whatever reason.
If the second commit is deemed controversial it can be shelved and current behavior documented alone.