Change 6948 added support for sample datasets, but there's work left to do:

1. ALTER TABLE t ADD SAMPLE ( ... );
2. ALTER TABLE t DROP SAMPLE ( ... );
3. CREATE [OR REPLACE] VIEW v SAMPLE ( ... );
4. Give error if sample names within a table are not unique.
5. Give error if a sample S for table T does not contain a column for each column in T.
6. Give error if a column CS in sample S for table T does not have the same datatype as the corresponding column CT in T. Implicit coercions are allowed. Also nullability.
7. If table T has a sample S, user cannot drop S. With 'DROP s CASCADE' keyword, the sample is removed from T but T is not dropped.
8. Include samples in generated DDL. Also, add the sample relationship to the partial order, so that sample tables are exported before the tables which depend on them.
9. Detect cycles in sample tables (E.g. x samples from y samples from x) and issue a specific error. Doesn't matter whether error is at DDL time or at runtime -- runtime is probably easier.