Entity-escaping in textarea fields happens for UTF8 strings

Testcase:

1. Create new note for any contact containing string "łąś"
2. Edit newly created note (diacritics are displayed as ���)
3. Save newly created note (diacritics are saved as ���)

Reproduced on http://sandbox.civicrm.org/

First investigation: Something treats this as ISO-8859-1 data and happily replaces the bytes by their HTML entities - the page's source is actually ���

Second investigation: Tried this for a couple of of other fields, and it seems this happens for textareas only, but all of them (tried Activities, Grants).