v0.7.0

fixes

1. SetBuilder missing terminal method — Added Set() Set[T] terminal method that returns the built set and invalidates the builder. Renamed the old Set(val T) to Add(val T) to avoid the name collision.
2. Value receivers on SetBuilder/SortedSetBuilder — Changed all methods on both builders to pointer receivers. Changed SetBuilder.s from Set[T] to *Set[T]. Renamed SortedSetBuilder.Set(val T) to Add(val T) for consistency. Added assert guards on all methods. Added invalidation and builder tests.
3. BatchMapBuilder.Flush inconsistent dedup — Fixed the large-buffer dedup path to iterate in reverse and then reverse the result, ensuring last-write-wins semantics consistent with all other code paths. Added TestBatchMapBuilder_LastWriteWins with four sub-tests.
4. StreamingMapBuilder unusable external API — Replaced the unexported mapEntry[K,V] parameter type with iter.Seq2[K,V] on all three methods (Stream, Filter, Transform). Removed the MapEntry interface and KV struct. Added/updated tests using maps.All.
5. t.Fatalf in goroutines — Replaced all t.Fatalf calls inside goroutines with t.Errorf + return. go vet now passes cleanly.
6. Queue complexity claim incorrect — Rewrote the entire Queue implementation to use an internal consList[T] (singly-linked list) instead of *List[T], achieving true amortized O(1) enqueue/dequeue. Rewrote QueueIterator to walk cons-list nodes. Restored the O(1) claim in docs. Added 6 new tests (mixed ops, snapshot isolation, builder, empty ops, etc.).
7. Unused mutable parameter in toTrie — Removed the parameter and updated both call sites.

updates

1. Oversized slice nodes from batch flush — Documented the intentional trade-off: batch-built lists may exceed listSliceThreshold, deferring the trie conversion to the first subsequent mutation.
2. Identity hash for small integers — Documented the trade-off on hashUint64: identity hashing causes HAMT root collisions for sequential keys, but applying the finalizer unconditionally would break hash stability.
3. Update to go 1.26