Follow-up from "gdbusmessage: Limit recursion of variants in D-Bus messages"

The following discussion from !1201 (merged) should be addressed:

• @smcv started a discussion: (+1 comment)

  I think it would probably be a better test if the message was valid, other than in the limited ways in which we want it to be invalid? (But I realise that's going to be annoying to do, so this is not a veto.)