direct_consumer: clarity fixes

The results of partitioning were incorrectly named in
filter_stale_subscriptions.

Fix the names and additionally pull the iterators from the subspan
rather than doing wasteful (and dangerous) iterator math.

Author: joe-redpanda

src/v/kafka/client/direct_consumer/direct_consumer.cc

@@ -81,6 +81,7 @@ direct_consumer::fetch_next(std::chrono::milliseconds timeout) {
     } catch (ss::condition_variable_timed_out&) {
         co_return chunked_vector<fetched_topic_data>{};
     }
+    // fallthrough loop case
     co_return chunked_vector<fetched_topic_data>{};
 }
 
@@ -90,7 +91,7 @@ void direct_consumer::filter_stale_subscriptions(
     for (auto& topic_data : responses_to_filter) {
         auto& partition_data = topic_data.partitions;
 
-        auto non_stale_subsegment = std::ranges::partition(
+        auto stale_subsegment = std::ranges::partition(
           partition_data,
           [this, &topic_data](const fetched_partition_data& data) mutable {
               auto maybe_current_subscription_epoch = find_subscription_epoch(
@@ -113,20 +114,18 @@ void direct_consumer::filter_stale_subscriptions(
               return !is_stale;
           });
 
-        auto erase_iterator = partition_data.end()
-                              - non_stale_subsegment.size();
-
+        // erase everything that is stale
+        auto erase_iterator = stale_subsegment.begin();
         partition_data.erase_to_end(erase_iterator);
     }
 
     // remove newly empty topics
-    auto non_empty_subsegment = std::ranges::partition(
+    auto empty_subsegment = std::ranges::partition(
       responses_to_filter, [](const fetched_topic_data& topic_data) {
           return !topic_data.partitions.empty();
       });
 
-    responses_to_filter.erase_to_end(
-      responses_to_filter.end() - non_empty_subsegment.size());
+    responses_to_filter.erase_to_end(empty_subsegment.begin());
 }
 
 void direct_consumer::update_offsets_filter_redundant_fetches(