Stop cloning index input when loading NRTSuggester (#14271)

Cloning the index input is not necessary when loading the NRTSUggester. Also,
it causes a bug as the clone gets closed right before the suggester instance
gets returned, which makes it unusable in the later lookup method which slices
the index input while it's already closed.

This was not discovered so far due to a testing gap that is fixed by #14270

Author: javanna

lucene/CHANGES.txt

@@ -130,6 +130,7 @@ Bug Fixes
 * GITHUB#14008: Counts provided by taxonomy facets in addition to another aggregation are now returned together with
   their corresponding ordinals. (Paul King)
 * GITHUB#14027: Make SegmentInfos#readCommit(Directory, String, int) public (Luca Cavanna)
+* GITHUB#14271: Stop closing index input when loading NRTSuggester (Luca Cavanna)
 
 Build
 ---------------------

lucene/suggest/src/java/org/apache/lucene/search/suggest/document/CompletionsTermsReader.java

@@ -72,10 +72,8 @@ public final class CompletionsTermsReader implements Accountable {
    */
   public synchronized NRTSuggester suggester() throws IOException {
     if (suggester == null) {
-      try (IndexInput dictClone = dictIn.clone()) { // let multiple fields load concurrently
-        dictClone.seek(offset);
-        suggester = NRTSuggester.load(dictClone, fstLoadMode);
-      }
+      IndexInput indexInput = dictIn.slice("NRTSuggester", offset, dictIn.length() - offset);
+      suggester = NRTSuggester.load(indexInput, fstLoadMode);
     }
     return suggester;
   }