Note the problem seems to be with multipart messages where they are yenced,
single part yencs are not a problem.

There seems a new way grab things from the cache in pan, it has major
problem, as the files are no longer loaded into memory (which would have
just caused a memory leak) there is now a FD leak and after decoding things
are not completely cleaned up.

This leak is serious as its the FDs for the cached messages that are
leeking that means only less than 1024 parts can be downloaded before
running into major problems. Which make pan unusable.

This system of grabbing the cached objects using and FD is limited, as the
FDs are all allocated "up front" this means that, only messages with less
than 1024 parts can be downloaded, I have been able to download messages
with more than 6000 parts before!!!!

Ideally, when dealing with a multipart message, the cached files would be
loaded serially into memory or by fd and discarded after pre-process as
this would minimise the resource use.