Bugs in python backend

I am creating an issue rather than a PR because I have not been able to create a working Odb or Refdb backend, but I have found several bugs. Please let me know if I am completely misunderstanding this code.

I believe the issues I cannot get past come from threading in libgit2. I have found that, at times, calling threading.get_ident will cause a segfault. I have tried wrapping all C->python code in PyGILState_Ensure and PyGILState_Release, but this did not work.

I have found the following possible errors:

• https://github.com/libgit2/pygit2/blob/e012bb5bc9e0bdb68ac79f9c12e3f59e3513328a/src/odb_backend.c#L244 Calls exists_prefix not refresh.
• https://github.com/libgit2/pygit2/blob/e012bb5bc9e0bdb68ac79f9c12e3f59e3513328a/src/refdb_backend.c#L134 glob may be null
• https://github.com/libgit2/pygit2/blob/e012bb5bc9e0bdb68ac79f9c12e3f59e3513328a/src/refdb_backend.c#L134 git_reference* _ref will double free
• https://github.com/libgit2/pygit2/blob/e012bb5bc9e0bdb68ac79f9c12e3f59e3513328a/src/refdb_backend.c#L409 PyIter_Check should be a check for __iter__ not if the object is currently an iterator.

In my experience, problems like this mean that I am misusing something. Let me know if these appear to be real bugs and I will see if I can at least create test cases to expose them. I don't seem to have a good grasp on the threading segfaults though.

Start by reading the cited sections of src/odb_backend.c and src/refdb_backend.c, then reproduce the Odb and Refdb backend problems, including the reported libgit2 threading and GIL-related segfault. Done means each suspected bug is confirmed or ruled out with reproducible coverage and the findings are resolved or clearly documented.