This allows for efficient reading of many references and their targets,
without incurring the overhead of lookup_reference() (which stats for
a loose ref and then reads packed-refs) which can be expensive on NFS
with thousands of refs.
Use the right namespace so Refspec and TransferProgress show up. These
classes are not instantiated by the user, so they were not promoted to
the pygit2 module. Use autoclass so it lists everything for us.
Also use autoclass for the reflog. We had a difference is oid vs id
between the docs and the code.
Now Repository.head behaves like libgit2's git_repository_head, it
returns the resolved reference. These two lines are equivalent:
ref = repo.head
ref = repo.lookup_reference('HEAD').resolve()
Before it returned a commit.
