MIN()/MAX() over BIGINT UNSIGNED loses the UNSIGNED flag when converted to a float

CREATE TABLE t (sh BIGINT UNSIGNED, g INT);

INSERT INTO t VALUES (18446744073709551588, 1);   -- 2^64 - 28

SELECT MIN(sh)                  FROM t;   -- 18446744073709551588        correct

SELECT CAST(sh AS DOUBLE)       FROM t;   -- 1.8446744073709552e+19      correct (no aggregate)

SELECT CAST(MIN(sh) AS DOUBLE)  FROM t;   -- -28.0                       WRONG

SELECT MIN(sh) + 0e0            FROM t;   -- -28.0                       WRONG (no CAST needed)

SELECT SQRT(MIN(sh))            FROM t;   -- NULL   (expected 4294967296) WRONG