I edited my post after verifying the actual bytes, it turned out to be slightly more complicated.

The double-encoding path gets you there too: the original UTF-8 \xE2 \x80 \x99 mis-decoded as iso-8859-1 or Windows-1252 and saved back as UTF-8 gives \xC3 \xA2 \xC2 \x80 \xC2 \x99, which in Windows-1252 renders as ’. A WYSIWYG cleanup replacing that mojibake with the Windows-1252 ' (byte 0x92) and saving back as UTF-8 gets you to \xC2 \x92 on disk.

Edit: Although maybe that's not the most parsimonious explanation.