ha_tina (CSV) deletes wrong records

--source include/have_csv.inc

CREATE TABLE t1 (v INT NOT NULL) ENGINE=CSV;

INSERT INTO t1 VALUES (1),(2),(3);

DELETE FROM a1, a3 USING t1 AS a1 JOIN t1 AS a3 WHERE a1.v=2 AND a3.v=1;

--echo # asked to remove v=1 and v=2, so v=3 must survive

SELECT * FROM t1 ORDER BY v;

DROP TABLE t1;

Some pre-analysis:
ha_tina backing CSV deletes the records physically from the file on every delete_row.
position() literally writes a file offset.
rnd_pos locates a record by than file offset

Scenario:
multi-delete records all the positions (=file offsets) to delete.
Then it deletes by those offsets in a single pass.
The offsets are not reverse-sorted, so when off1 < off2 and off1 is deleted first, off2 will point to a wrong record.

The fix is either of three:
a) Implement delete-marking in tina/csv
b) Reverse-sort the position buffer.
c) delete immediately after locating, skipping the buffer