Improve performance of scans on tablets with large blocks of deleted rows

Scanning tablets where many of the rows have been deleted is often very slow, since the deletes need to be applied as deltas. It's possible to speed this up in cases where an entire block has been deleted by
1. skipping column materialization for a block if all rows have been deleted
2. skipping reading the selection vector for each column if we know the entire block is deleted and thus contains no results