mastoapi allocates too much memory
see Potential memory leak? Every four hours, the se... (#805 - closed) for the investigation
apparently fetching 40 notes from the global timeline via the mastoapi layer is enough to crash some sharkey servers
from the code I can see that we'd have 3 copies of the notes in ram at the same time (the response from misskey api, the mastodon-shaped converted data, and the json-serialised version of that inside reply.send), but that should still not be enough