sql: EPSC oracle should detect row-order divergences when ORDER BY is present

**Is your feature request related to a problem? Please describe.**
The EPSC oracle compares rows between the simple and prepared query paths *modulo order* (see the row comparator and the "rows agree modulo order" test case). That is correct for queries with no ordering guarantee, but it means an explicit `ORDER BY` that the prepared path returns in a different order than the simple path goes undetected — a genuine ordering bug would be missed.

**Describe the solution you'd like**
When the statement's output order is fully determined (top-level `ORDER BY`, possibly with `LIMIT`), compare order-sensitively; otherwise keep the order-insensitive comparison.

**Describe alternatives you've considered**
Always comparing order-sensitively — rejected because queries without an ordering guarantee may legitimately return rows in different orders, producing false positives.

**Additional context**
The crux is deciding when order is actually determined: `ORDER BY` on a non-unique key still permits ties in any order, so strict ordering should only be enforced to the extent the sort key makes the result deterministic (e.g. compare the projected `ORDER BY` key sequence, or only enforce strict order when the sort key is provably unique). Follow-up raised during review of the EPSC oracle work.

Jira issue: CRDB-66115

Start with the EPSC oracle's row comparator and the "rows agree modulo order" test case mentioned in the issue. Determine how the statement's top-level ORDER BY, LIMIT, projected sort keys, and key uniqueness can establish deterministic output order. Done means ordered queries detect relevant divergences while unordered queries and legitimate ties remain order-insensitive, with tests covering both behaviors.

Written by the indexing model from the issue text.