fix: disable overflow passthrough for distinct aggregates

COUNT(DISTINCT) and similar distinct aggregates produce per-row
intermediate state when convert_to_state is called. In overflow
mode this turns 100M rows into 100M single-value state objects
that the downstream must merge — a 12x regression on Q9.

Fix: skip overflow passthrough when any aggregate is distinct.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: Dandandan

datafusion/physical-plan/src/aggregates/row_hash.rs

@@ -674,10 +674,14 @@ impl GroupedHashAggregateStream {
         };
 
         // Overflow passthrough requires convert_to_state support
-        // (same requirement as skip_aggregation_probe)
+        // (same requirement as skip_aggregation_probe) and no distinct
+        // aggregates (convert_to_state for distinct produces per-row
+        // state objects that are catastrophically expensive to merge).
+        let has_distinct = aggregate_exprs.iter().any(|e| e.is_distinct());
         let overflow_passthrough_max_table_size = if agg.mode == AggregateMode::Partial
             && matches!(group_ordering, GroupOrdering::None)
             && skip_aggregation_probe.is_some()
+            && !has_distinct
         {
             context
                 .session_config()