Cleanup use of federation metadata in client / avoid race conditions

[bbockelm]

When cleaning up #721, it became obvious that the client code structure is wrong in its handling of metadata.

The client uses the global discovery metadata. However, the "correct" director, for example, may change URL-by-URL -- and if we are downloading multiple URLs in parallel, there's going to be race conditions setting/unsetting the global director location for each URL we parse.

Instead of reaching into the global we should create a URL-like structure that holds the metadata along with the object name. We should also keep a ttlcache of metadata for different federations to prevent frequent lookups if we are parsing many URLs within the process.