Parallel pull for ssh-pull

I think I've now got the parallel pull use in ssh-pull to the point where 
it could be useful to post. Similar stuff should work for http-pull (where 
it will probably be more interesting), but I have to read more libcurl 
documentation.

Initial results on ssh-pull are encouraging: on my local ethernet, I moved 
the current git repository in 16 seconds, unpacked and with the push 
side acting dumb. I think it was requesting objects about 6 ahead of where 
it was reading them. I didn't test how much time the pipe spent empty in 
the final version; when I was printing debugging messages, the pipe was 
often empty, but it also took 3 times as long, and was therefore probably 
blocking on debugging output, not the network.

This work is based on some now in -pu; what should I base my patches on? I 
would ideally like to add a function to one of the patches in -pu and fix 
a subtle bug in the other, in addition to further patches to actually use 
the feature in ssh-pull.

	-Daniel
*This .sig left intentionally blank*