Sparse without dups unordered uses indexreader (#3347) (#3348)

This fixes a mismatch between the serialization path and the query
strategy decision code that resulted in the serialization path
incorrectly trying to load the base reader state instead of the indexed
reader state.

tiledb/sm/serialization/query.cc

@@ -1271,8 +1271,9 @@ Status query_to_capnp(
       }
     } else if (
         query.use_refactored_sparse_global_order_reader() && !schema.dense() &&
-        (layout == Layout::GLOBAL_ORDER ||
-         (layout == Layout::UNORDERED && query.subarray()->range_num() <= 1))) {
+        (query.condition() == nullptr || query.condition()->empty()) &&
+        (layout == Layout::GLOBAL_ORDER || layout == Layout::UNORDERED) &&
+        query.subarray()->range_num() <= 1) {
       auto builder = query_builder->initReaderIndex();
       auto reader = (SparseGlobalOrderReader*)query.strategy();
 
@@ -1801,7 +1802,7 @@ Status query_from_capnp(
   // heterogeneous coordinate changes
   if (type == QueryType::READ) {
     if (query_reader.hasReaderIndex() && !schema.dense() &&
-        layout == Layout::GLOBAL_ORDER) {
+        (layout == Layout::GLOBAL_ORDER || layout == Layout::UNORDERED)) {
       // Strategy needs to be cleared here to create the correct reader.
       query->clear_strategy();
       RETURN_NOT_OK(query->set_layout_unsafe(layout));