Cyrillic names arrived as question marks and were stored that way.
Upload: BytesParser defaults to the compat32 policy, which wraps a
non-ASCII Content-Disposition in a Header with the unknown-8bit
charset, so get_filename() returned one replacement character per
UTF-8 byte and the mangled name was written to disk. Parse with
policy=HTTP, which decodes headers as UTF-8.
Download: send_header() encodes latin-1 strict, so a Cyrillic name
raised UnicodeEncodeError. Emit both Content-Disposition forms per
RFC 5987 — an ASCII fallback plus percent-encoded filename*.
eberkheev
