downtime should not result in missing content when the sending instance is lemmy 0.19.0 or newer. 0.19.0 introduced a persistent federation queue in lemmy, which means it will retry sending the same stuff until the instance is available. depending on the type of down, it can also be possible that there is a misconfiguration (e.g. "wrong" http status code on a maintenance page) that could make the sending instance think it was successfully sent. if the sending instance was unreachable (timeout) or throwing http 5xx errors, everything should be preserved.
we are planning to post an announcement about the current situation with lemmy updates and our future plans in the coming days, stay tuned for that. you can find some info in my comment history already if you are curious.
there is indeed a cutoff. there is exponential delay for retrying and at some point lemmy will stop trying until it sees the instance as active again.
there is also a scheduled task running once a week that will delete local activities older than a week. downtimes of a day or two can generally be easily recovered from, depending on latency it can take a lot more time though. if an instance is down for an extended time it shouldn't expect to still get activities from the entire time it was offline.