[kvlonge]

For C's form linking. The clause classifier gave GROUP BY ALL, ORDER BY ALL and WITH RECURSIVE only their plain clause keys, and gave a SAMPLE on a table reference no key at all, because only Select.sample was read. So group-by.all, order-by.all, with.select.syntax and sample could never be measured.

Change (crates/sqlquack-core classification): new keys, each emitted alongside the existing clause key:

• clause:group_by_all when GroupBy.all == Some(true);
• clause:order_by_all when OrderBy.all is set, on SELECT and set operations;
• clause:with_recursive when With.recursive, on SELECT and set operations;
• clause:sample on a table reference with table_sample.

coverage/translation/form-links.toml links each one to its form. It's a fail-closed file: an unlinked key fails the measure. C adds the refusal links and the join_using correction afterwards.

Evidence

• A unit test in test_classify covers each key. sqlquack census shows a plain GROUP BY a ORDER BY a still emits only clause:group_by and clause:order_by.
• verify-recorded (main's cases, at 90f133d with the same converter): 8470 cases: 6780 passed, 994 diverged as expected, 660 refused, 36 typed, 0 pending, 0 unproven, 0 failed; 220 of 220 equivalences and rules verified.
• Translation measure against main: 0 regressions, 1 other change. sf.query.group-by.all goes from unmeasured to matching (4 recorded statements). The other three forms have no recorded statement on main yet; C's slices bring them. Baseline re-recorded.
• Conversion and refusal are unchanged: the keys are added to occurrences, and no rule or policy names them.