Revert "Disable concurrent search for filter duplicates" (#20915)

Concurrent search creates a new aggregator per slice, so there is no
concurrency problem here. The dedup logic is currently per-shard with no
reduce logic at the coordinator, so concurrent search does change that
to be per-slice, but does not fundamentally change the behavior. I'm
reverting this and will deal with test flakiness if it continues.

This reverts commit f46ce23.

Signed-off-by: Andrew Ross <andrross@amazon.com>

Author: andrross

CHANGELOG.md

@@ -31,7 +31,6 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 - Use env variable (OPENSEARCH_FIPS_MODE) to enable opensearch to run in FIPS enforced mode instead of checking for existence of bcFIPS jars ([#20625](https://github.com/opensearch-project/OpenSearch/pull/20625))
 - Update streaming flag to use search request context ([#20530](https://github.com/opensearch-project/OpenSearch/pull/20530))
 - Move pull-based ingestion classes from experimental to publicAPI ([#20704](https://github.com/opensearch-project/OpenSearch/pull/20704))
-- Disable concurrent search for filter duplicates in significant_text ([#20857](https://github.com/opensearch-project/OpenSearch/pull/20857))
 
 ### Fixed
 - Relax index template pattern overlap check to use minimum-string heuristic, allowing distinguishable multi-wildcard patterns at the same priority ([#20702](https://github.com/opensearch-project/OpenSearch/pull/20702))

server/src/main/java/org/opensearch/search/aggregations/bucket/terms/SignificantTextAggregatorFactory.java

@@ -315,9 +315,6 @@ public void close() {
 
     @Override
     protected boolean supportsConcurrentSegmentSearch() {
-        // The underlying structured used to find duplicate byte sequences (DuplicateByteSequenceSpotter)
-        // is stateful and not thread-safe, so disable concurrent search if this particular feature is
-        // being used.
-        return filterDuplicateText == false;
+        return true;
     }
 }