* core/state/snapshot, ethdb: track deletions more accurately

* core/state/snapshot: don't reset the iterator, leveldb's screwy

* ethdb: don't mess with the insert batches for now