Improve attachment filename extraction

[njohner]

We were already handling the case where filename returned by msg.get_filename() returned a cropped version of the filename. In that case we took the filename from the last filename parameter of the content-disposition headers. It now turns out that sometimes msg.get_filename() returns a normalized version of the filename, e.g. without umlauts. This is also not what we want and using the last filename parameter of the content-disposition headers would also fix the issue. So it seems we can always take the last one when there are several.

Note that I changed/fixed a few other things:

• We pretended to look at the content-disposition header when we were actually looking at the content-type header.
• We now search for the "filename" parameter in the header case-insensitive. This is also what python's Message.get_param does.

For https://4teamwork.atlassian.net/browse/CA-5456

[phgross]

👍 LGTM