Dynamically grow hash in linear counting in HLL (elastic#142047) (elastic#142347)

This change dynamically grows the hash inside each linear counting
instance independently to address high memory usage in count_distinct,
especially in cases with a large number of groups (1M+) that each
contain a small number of distinct values (<3000). Instead of eagerly
allocating 16K for each grouping, this dynamic approach serves as an
immediate optimization. The long-term plan, however, is to create a new,
separate HLLState for ES|QL that is more friendly to Block/Vector and
ES|QL aggregations.

Closes elastic#41847
Relates elastic#142333

---------
Co-authored-by: Matt <matthew.alp@elastic.co>
Co-authored-by: Nhat Nguyen <nhat.nguyen@elastic.co>

(cherry picked from commit a0b6bb6)

Co-authored-by: Matt <matthew.alp@elastic.co>

Author: dnhatn

docs/changelog/142047.yaml

@@ -0,0 +1,6 @@
+area: ES|QL
+issues:
+ - 41847
+pr: 142047
+summary: Dynamically grow hash in linear counting in HLL
+type: bug

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

@@ -96,5 +96,24 @@ public interface HashesIterator {
          * @return the current value of the counter.
          */
         int value();
+
+        HashesIterator EMPTY = new EmptyIterator();
+    }
+
+    private static class EmptyIterator implements HashesIterator {
+        @Override
+        public int size() {
+            return 0;
+        }
+
+        @Override
+        public boolean next() {
+            return false;
+        }
+
+        @Override
+        public int value() {
+            throw new IllegalStateException("empty iterator");
+        }
     }
 }