This issue has been resolved.
The cause of the issue was a queue that clusters get put into after they identify with Aether based on the max concurrency returned by Discord in /gateway/bot to prevent too many shards starting at once.
This queue took too long for some clusters to be assigned an id and shards so they disconnected and entered a race condition that allowed two of them be assigned the same id and shards, causing the double responses and the shard it was supposed to be given was now stuck offline.
I noticed this issue pretty quickly but I have added more robust monitoring for this exact scenario to alert me almost immediately.