The dict of headers of request objects in urllib has its keys always
capitalized.
This causes the lookup to fail and overwrite the header. If for example
a Extractor tries to add a "User-Agent" header the internal
representation in the request object is "User-agent". The header is
therefore clobbered by the "User-Agent" in std_headers, because the
strings are not equal.
If one of the processors said the file should be kept, it wouldn't pay
attention to the response from the following processors. This was wrong if the
'keep_video' option was False, if the first extractor modifies the original file
and then we extract its audio we don't want to keep the original video file.