Add support for walking through shallow clones

Rugged doesn't support walking through the commits of partially cloned repositories:

In the shell do:

cd /tmp/
git clone --depth 1 git@github.com:eugenk/importing_owl.git

And then in the ruby console do:

r = Rugged::Repository.new('/tmp/importing_owl')
walker = Rugged::Walker.new(r)
walker.push(r.head.target.oid)
walker.each { |c| puts c.inspect }

which results in

Rugged::OdbError: Object not found - failed to find pack entry (c84d1be7573e2367e175b8675fcc347aaf13a27b)

where c84d1b is the parent commit oid.

Rescuing from Rugged::OdbError won't help because the walker raises this error as soon as the iterator starts walking - not when trying to find the missing parent commit.

This happens regardless of the sorting method (walker.sorting(Rugged::SORT_REVERSE) or walker.sorting(Rugged::SORT_TOPO))

If the oldest existing commit is a merge commit, then the error is not being raised. However, absolutely nothing happens, i.e. no commits are being iterated over.

The expected behaviour is at least to run through the commits which actually exist and then raise an error.

Reproduce the shallow-clone case using Rugged::Repository and Rugged::Walker as shown, including both sorting modes and the merge-commit case. Start by tracing when walker iteration begins and how missing parent commits are handled. Done means locally available commits are yielded before the missing-parent error is reported.