I find those bugs less puzzling because the timeline makes sense. You're not logged in, there's a prompt, you're logged in. Obvious bug in the prompt, but the A happens before B happens before C order is there. Call comes in, audio is recorded, call is accepted is not the expected order. I could imagine a bug where declining the call still accepts the call, because that still obeys the proper ordering, but this bug does not.
I don’t use face time often, but isn’t there a "preview" of the camera feed during the prompt? I guess so that the user can check if he’s looking decent before engaging the call.
The bug could then be that the feed is sent over the call too early instead of being used solely for this local feedback.
From the details, it sounds like "adding" the caller to the call before the call recipient accepts probably puts it in a weird state. Could be some kind of off-by-one error, where for some purposes Participant 2 is the caller and for some it's the recipient.