As for autofill memory limitations, this is largely an implementation detail: just process data stream in small chunks, that's it.
Unfortunately, KeePassium's data layer was designed in the times of iOS 11, before AutoFill became a thing. So I chose the easier path of loading and processing the whole file at once. This made sense for 10-20 MB databases on iPhones with 2 GB of RAM. By the time the mistake became obvious, it was much harder to switch to streamed processing, especially with a long queue of lower-hanging feature requests.
Unfortunately, KeePassium's data layer was designed in the times of iOS 11, before AutoFill became a thing. So I chose the easier path of loading and processing the whole file at once. This made sense for 10-20 MB databases on iPhones with 2 GB of RAM. By the time the mistake became obvious, it was much harder to switch to streamed processing, especially with a long queue of lower-hanging feature requests.