Fetch tree and blob contents lazily

[danports]

The current eager fetching approach works fine for smaller repositories, but has disastrous performance for even medium-sized repositories when you don't need to clone the entire repository.

I needed this for https://github.com/danports/amber/issues/22 and originally implemented it in https://github.com/danports/computercraft-github/commit/04b44aade807c9372d8b377756370df7a4ab8986.