lru.nx

lru.nx -- Least-Recently-Used cache. Composes map.nx (key -> node) + dlist.nx (recency order). Classic design, same shape as the LRU in every modern kernel page cache + every language std lib's LinkedHashMap: - get(k) -> value + moves node to front - put(k, v) -> insert or update + moves node to front, evicts tail if over capacity This is the first module composing two runtime data structures end-to-end. Proves the shelves interlock cleanly. Storage per entry: DNode with data pointer holding a LRUEntry struct {key, val}. Map stores key -> node pointer so get() is O(1) amortised. Use cases: compiled-template cache (nishi-pages), route table in HTTP router, DNS resolver cache, prepared-SQL cache, image thumbnail cache. Invariants: L1 size <= capacity at all times (enforced by eviction on insert). L2 Map and dlist stay in sync: every map entry value is a live DNode in the dlist; every live dlist node has an entry in the map. L3 get(missing) returns -1 without mutating state. L4 Capacity of 0 is legal (no entries ever admitted).