const MAX_CACHED_HINT_INSTANCES: usize = 16;
We keep this small to enable a simple LRU cache with a linear search. Regenerating hinting data is low to medium cost so it’s fine to redo it occasionally.