MDEV-19600: The optimizer should be able to produce rows=1 estimate for unique index with NULLable columns

Modify best_access_path() to produce rows=1 estimate for null-rejecting
lookups on unique NULL keys.

mysql-test/main/subselect_sj.test

@@ -2285,16 +2285,66 @@ INSERT INTO t1 VALUES
   (1,2),(2,1),(3,3),(4,2),(5,5),
   (6,3),(7,1),(8,4),(9,3),(10,2);
 
-CREATE TABLE t2 ( c INT, d INT, UNIQUE KEY(c) );
+CREATE TABLE t2 ( c INT, d INT, KEY(c) );
 INSERT INTO t2 VALUES
   (1,2),(2,1),(3,3),(4,2),(5,5),(6,3),(7,1);
 
+analyze table t1,t2;
+explain
+SELECT a, b, d FROM t1, t2
+WHERE ( b, d ) IN
+  ( SELECT b, d FROM t1, t2 WHERE b = c );
 SELECT a, b, d FROM t1, t2
 WHERE ( b, d ) IN
   ( SELECT b, d FROM t1, t2 WHERE b = c );
 
 DROP TABLE t1, t2;
 
+--echo # Another testcase for the above that still uses LooseScan:
+
+create table t0(a int primary key);
+insert into t0 values (0),(1),(2),(3),(4),(5),(6),(7),(8),(9);
+
+create table t10(a int primary key);
+insert into t10 select A.a + B.a* 10 + C.a * 100 from t0 A, t0 B, t0 C;
+
+create table t1 (
+   pk int primary key auto_increment,
+   kp1 int,
+   kp2 int,
+   filler char(100),
+   key (kp1, kp2)
+);
+
+# 10 groups, each has 10 elements.
+insert into t1 (kp1, kp2, filler) 
+select 
+  A.a, B.a, 'filler-data'
+from t0 A, t0 B;
+
+create table t2 (a int, filler char(100), key(a));
+
+create table t3 (a int);
+insert into t3 values (1),(2);
+
+insert into t2 
+select (A.a+1000*B.a)/20, 'filler_data' from t10 A, t0 B;
+
+analyze table t1,t2,t3;
+delete from t1 where kp2 in (1,3);
+
+--echo # Ref + LooseScan on t1:
+explain select sum(t2.a)
+from t2,t3
+where (t3.a,t2.a) in (select kp1,kp2 from t1,t0 where t0.a=2);
+
+select sum(t2.a)
+from t2,t3
+where (t3.a,t2.a) in (select kp1,kp2 from t1,t0 where t0.a=2);
+
+drop table t0,t10;
+drop table t1,t2,t3;
+
 --echo #
 --echo # BUG#920713: Wrong result (missing rows) with firstmatch+BNL, IN subquery, ...
 --echo #