Disable parallel collection in top hits aggregator

TopHitsAggregator calls TopFieldCollector#populateScores from each slice, which in turn
relies on createWeight calls against the searcher. This means that the
createWeight call gets parallelized with Lucene 9.9, but it should not
as we are already executing multiple slices concurrently. We have
protecions in place in Lucene to avoid this problem, but we have to
adapt our execution code in ContextIndexSearcher accordingly.

The goal is to utilize plain TaskExecutor from Lucene and avoid code
duplication. In the meantime, we can disable concurrency for top hits
aggregations as it causes actual issues currently in the lucene_snapshot
branch.

server/src/main/java/org/elasticsearch/search/aggregations/metrics/TopHitsAggregationBuilder.java

@@ -921,4 +921,9 @@ public String getType() {
     public TransportVersion getMinimalSupportedVersion() {
         return TransportVersions.ZERO;
     }
+
+    @Override
+    public boolean supportsParallelCollection() {
+        return false;
+    }
 }

server/src/test/java/org/elasticsearch/search/aggregations/metrics/TopHitsTests.java

@@ -183,4 +183,8 @@ protected void assertToXContentAfterSerialization(TopHitsAggregationBuilder orig
             XContentType.JSON
         );
     }
+
+    public void testSupportsParallelCollection() {
+        assertFalse(createTestAggregatorBuilder().supportsParallelCollection());
+    }
 }