small patch for async writes

Hi,

I tried enabling the ASYNC_WRITE define in src/fs/aufs/store_asyncufs.h
and found that the the cache log was filling up with "this be aioCancel.
Danger ahead!" errors. The squid-users list says that these errors are
harmless, but I found that they were not, as objects were not being fully
written to disk. The attached patch seems to fix this by fixing what
seems to be a logic flaw.

The only thing I'm not sure about is what will happen if another request
wants to open this file while flags.close_request is set to 1. This would
be a problem, the solution to this would be to get the process to block
until all io has finished on this filedescriptor.

Steven

ps. I've haved together the visolve realtime signal code with the squid
3.0 epoll code to make a version of squid 2.5 that supports epoll. It
overrides the select and poll methods, but it does scale very well.