MDEV-26709: JSON histogram may contain bucketS than histogram_size allows

When computing bucket_capacity= records/histogram->get_width(), round
the value UP, not down.

mysql-test/main/statistics_json.test

@@ -261,3 +261,19 @@ DROP TABLE t1;
 --echo #
 select variable_comment from information_schema.system_variables where VARIABLE_NAME='HISTOGRAM_TYPE';
 
+--echo #
+--echo # MDEV-26709: JSON histogram may contain bucketS than histogram_size allows
+--echo #
+create table t1 (a int);
+insert into t1 values (1),(3),(5),(7);
+insert into t1 select 2 from seq_1_to_25;
+insert into t1 select 4 from seq_1_to_25;
+insert into t1 select 6 from seq_1_to_25;
+
+set histogram_size=4, histogram_type=JSON_HB;
+analyze table t1 persistent for all;
+
+select histogram from mysql.column_stats where table_name = 't1';
+
+drop table t1;
+

sql/opt_histogram_json.cc

@@ -100,7 +100,7 @@ class Histogram_json_builder : public Histogram_builder
   /*
     Number of rows that we intend to have in the bucket. That is, this is
 
-      n_rows_in_table / histo_width
+      n_rows_in_table / hist_width
 
     Actual number of rows in the buckets we produce may vary because of
     "popular values" and rounding.
@@ -129,7 +129,14 @@ class Histogram_json_builder : public Histogram_builder
                          ha_rows rows)
     : Histogram_builder(col, col_len, rows), histogram(hist)
   {
-    bucket_capacity= records / histogram->get_width();
+    /*
+      When computing number of rows in the bucket, round it UP. This way, we
+      will not end up with a histogram that has more buckets than intended.
+
+      We may end up producing a histogram with fewer buckets than intended, but
+      this is considered tolerable.
+    */
+    bucket_capacity= round(rows2double(records) / histogram->get_width() + 0.5);
     if (bucket_capacity == 0)
       bucket_capacity= 1;
     hist_width= histogram->get_width();