- ASCII fast path in bytes_to_string (~13% CPU reduction)
- GrepFilter fill_buf zero-copy path (~8% wall-clock improvement)
- split_off instead of drain().collect() in extract_complete
- Reduce address String cloning in MessageIterator
- Add --no-grep-filter CLI flag to skip GrepFilter overhead
- Document GrepFilter overhead for library consumers
- Fix remaining bytes not marked as skipped on EOF in FrameIterator
- Expand design rationale with performance and origin story