parse commit marks more leniently

parse commit marks more leniently

Older versions of git-log would print marked commits (like
boundary commits) like this:

  commit -bc911786d75680d0fa28ac81c846775f6a7c894f

However, since git v1.7.5, it now outputs:

  commit - bc911786d75680d0fa28ac81c846775f6a7c894f

As a result, we pass the extra space into `main_register_commit`,
which ends up copying the null sha1 as the commit id. As a result,
we would fail to match it with a later call to `main_add_commit`,
and would omit the marked commit from the graph.

Since we do not know which git version we are using, let's cover
both cases by simply skipping past any non-alphanumeric content,
including spaces and marks, which should put as at the commit id.

Note that we can just drop the "is_boundary" check in our loop,
as it is only a special case of "!isalnum". We need the latter to
also move past --left-right marks.