fix: add statement_timeout and lock_timeout to mirror_primary_tables to prevent FDW zombie locks

Add SET LOCAL timeouts to refresh_tables() to prevent FDW zombie
connections from holding ACCESS EXCLUSIVE locks on mirrored tables
indefinitely. This fixes a v0.42.0 regression where stuck FDW queries
during mirror_primary_tables() cascade into full system lock.

- statement_timeout=60s: kills stuck FDW queries (normally <5s)
- lock_timeout=10s: prevents TRUNCATE stampede from multiple nodes

Both are SET LOCAL (transaction-scoped), zero impact on other queries.

Author: suntzu93

store/postgres/src/primary.rs

@@ -2218,6 +2218,14 @@ impl Mirror {
             .await
         }
 
+        // Prevent FDW zombie connections from holding ACCESS EXCLUSIVE locks
+        // indefinitely. lock_timeout prevents TRUNCATE stampede when multiple
+        // nodes mirror simultaneously. statement_timeout kills stuck FDW
+        // queries. SET LOCAL scopes these to this transaction only.
+        conn.batch_execute("SET LOCAL statement_timeout = '60s'; SET LOCAL lock_timeout = '10s';")
+            .await
+            .map_err(StoreError::from)?;
+
         // Truncate all tables at once, otherwise truncation can fail
         // because of foreign key constraints
         let tables = Self::PUBLIC_TABLES