PackageIndex handles serializing and deserializing the package index,

which points to the main packages we keep in the package repository.
The intent for now is to show only the main packages (ReviewBoard,

RBTools, etc.) that we want quick information on. Consumers of

PackageIndex will know, with a single file fetch, the latest versions of

the packages and what package names they can match locally.
As part of this, the package objects can now properly load and chain.

The objects with manifest URLs now contain absolute manifest URLs as

well, built using urljoin() along with a base URL. This gives us the

ability to always grab the right file from any other file, starting at

the index onward.

Unit tests pass.
Used this manually in a test command I have, and was able to navigate

the entire tree of package data.