Is there a way to recover mangled encodings like "Ch�Teau"?
The words seem to contain that replacement question mark, '\uFFFD'. Which means data loss, but also a simple marker to find those values.
Make a table with as primary key the wrong string. Repairing might be easier than thought. In the example the capital R after the qm instead of r might stem from word wise capitalisation. i?r is most likely i e-gauche r.
A regexp replace then will do.
However it might be easier to reimport a corrected database dump. The repair does not need to be in SQL.