In ucs2/utf16/utf32 documents, JSON_EXTRACT and JSON_CONTAINS_PATH do not find paths that JSON_VALUE finds

When the document argument is in a charset with mbminlen > 1, the path lookup of JSON_EXTRACT / JSON_CONTAINS_PATH fails, while JSON_VALUE on the very same arguments succeeds.

Reproduced on 10.11.19 (debug build):

SELECT JSON_EXTRACT(CONVERT('{"a":[1,2]}' USING ucs2), '$.a[1]'),               -- NULL

       JSON_CONTAINS_PATH(CONVERT('{"a":[1,2]}' USING ucs2), 'one', '$.a[1]'),  -- 0

       JSON_VALUE(CONVERT('{"a":[1,2]}' USING ucs2), '$.a[1]');                 -- 2

SELECT JSON_EXTRACT(CONVERT('{"a":1}' USING utf32), '$.a'),                     -- NULL

       JSON_CONTAINS_PATH(CONVERT('{"a":1}' USING utf32), 'one', '$.a');        -- 0

Expected: all three agree (2 / 1 / 2, and 1 / 1).

Related: MDEV-34173 (character_set_connection=ucs2) and MDEV-31539 (character_set_database=ucs2) cover the connection/database charset; here the document itself is in a wide charset and the three functions disagree with each other on identical arguments.