Make S3 and Hadoop catalogs (or layers) self contained

[lossyrob]

Currently, the path for a layer in either the S3 or Hadoop backends specifies a whole path. This means that, if you were to move the catalog around, the paths would become invalid. We should fix this so that at least the layers paths are relative to the catalog paths; a better fix for these backends (and the filesystem backend) is to make the layers entirely self contained, so moving one directory can result in a complete layer move (right now we have the attributes and the tiles separated).

[echeipesh]

Placing the attributes in the layer folder would simplify the data management a lot. We should keep already existing layers readable, which will be a little tricky.