The distinct count in merged statistics between full and partial statistics doesn't seem to be consistent with the distinct counts of the stats being merged for larger row counts.

Example:

SET enable_create_stats_using_extremes = on;
CREATE TABLE a (a PRIMARY KEY) WITH (sql_stats_automatic_collection_enabled = false) AS SELECT generate_series(0, 10000);
CREATE STATISTICS a_full ON a FROM a;
SHOW STATISTICS FOR TABLE a;

INSERT INTO a VALUES (10001);
CREATE STATISTICS a_partial ON a FROM a USING EXTREMES;
SHOW STATISTICS FOR TABLE a WITH MERGE;

Observe that the full stat estimates a distinct count of 9920, and the partial stat estimates 1 distinct value beyond the previous extremes. However, the merged stat only estimates 9803 (where 9921 is expected).

Jira issue: CRDB-40092