Cleaner Multi-Repo handling

Looks like Trac 0.12 will contain multi-repository support. If so, it might be better to treat each branch as a repository, and let Trac deal with identifying them. This suggests refactoring the code into one repository implementation for the current multi-branch approach, and another class for a single branch. The latter could be used to cache branch-related data for the duration of a request (obsoleting BranchCache), while the former could probably delegate much of its work to the latter and could itself concentrate on dealing with the branch part of revisions and with unversioned directory nodes.