Enable connection and dns caching using urllib instead of pycurl

Right now we default to using PyCurl because it can maintain a DNS cache as well as supporting keep-alive for requests. However, it is a python extension, rather than a pure python implementation.
So on a long download ^C doesn't work properly, and it requires more work to distribute compiled code.
A beta version is available at http://bazaar.launchpad.net/~bzr/bzr/bzr.urllib.keepalive.