Escape sequence for UTF-8 Character Literals?

I’ve a data set in a text file which uses the U+XXXX syntax for UCS
characters which I want to enter into a (utf8) db, using the actual
characters rather than the codepoint names.

The docs give the impression that eg E'\x91D1' ought to be the same
as '金', but my tests show that \x only accepts 2 hex digits.

Is doing the conversion client side the only way to do this?

I’m on 8.2, if the answer is version-dependent.

-JimC
-- 
James Cloos <cloos@jhcloos.com>         OpenPGP: 1024D/ED7DAEA6