An undesirable feature of the current cell implementation is that every unpacked cell contains both a value-oriented time window and an address-oriented aggregate time window. This means that first, it's easy to mistakenly use the wrong time window information, and second, the unpacked cells are larger than they need to be (specifically, removing one of the time window structure would get us under 128B).
While we're in there, it would be nice to simplify the cell unpacking code to reduce its footprint.