Scratches don't reset tick_deltas at end of tick

Scratches don't separate storage from tick_deltas -- which is fine, but it also means that scanners over scratches that don't need to be rescanned emit redundant tuples in subsequent ticks.

Fixing this probably requires first fixing #292, because it seems that scratches that feed channels depend on this bogus behavior to produce the correct output right now.
