fix(admin): guard null topK/threshold in KnowledgeBaseDocumentRetriever (#4736)

The public retrieve(Query) auto-unboxes searchOptions.getSimilarityThreshold()
in the filter and searchOptions.getTopK() in the limit, so it throws an NPE
on the first RAG-backed chat turn whenever the agent-level FileSearchOptions
leaves topK/similarityThreshold unset (the default). The private
retrieve(KnowledgeBase, Query) in the same class already null-checks both;
mirror that null-safety here.

When topK is null the merged results are returned uncapped (Long.MAX_VALUE)
and a null similarityThreshold skips the threshold filter — per-KB retrieval
already applied its own. Resolves the NPE reported in #4734.

Optional companion (not included): adding @Builder.Default to
FileSearchOptions.topK/similarityThreshold would make the documented defaults
of 3 / 0.2 actually apply — left as a maintainer decision on the intended
semantics.

Author: XianReallyHot-ZZH

spring-ai-alibaba-admin/spring-ai-alibaba-admin-server-core/src/main/java/com/alibaba/cloud/ai/studio/core/rag/retriever/KnowledgeBaseDocumentRetriever.java

@@ -96,10 +96,13 @@ public List<Document> retrieve(@NotNull Query query) {
 				documents.addAll(future.get(SEARCH_TIMEOUT, TimeUnit.SECONDS));
 			}
 
+			Integer topK = searchOptions.getTopK();
+			Float similarityThreshold = searchOptions.getSimilarityThreshold();
 			List<Document> results = documents.stream()
 				.sorted(Comparator.comparing(Document::getScore, Comparator.nullsLast(Comparator.reverseOrder())))
-				.filter(x -> x.getScore() != null && x.getScore() > searchOptions.getSimilarityThreshold())
-				.limit(searchOptions.getTopK())
+				.filter(x -> x.getScore() != null
+						&& (similarityThreshold == null || x.getScore() > similarityThreshold))
+				.limit(topK != null ? topK.longValue() : Long.MAX_VALUE)
 				.toList();
 
 			LogUtils.monitor("DocumentRetriever", "retrieve", start, SUCCESS, query.text(), results.size());