Make dimension attributes an array of objects, each mapping several logical model fields to source CSV columns

Adapted with slight modification from this comment by @pwalsh: https://github.com/openspending/fiscal-data-package/issues/48#issuecomment-158681781

Includes change of fields to attributes suggested here: https://github.com/openspending/fiscal-data-package/issues/93
Includes partial reversion from fields as array elements to fields as object elements: https://github.com/openspending/fiscal-data-package/issues/94

Reasoning for this change:

It is currently harder than it should be apply levels using the level keyword. That is, an implementer has to make a (somewhat arbitrary) choice as to what number to assign what level when all we want to do is determine relative position: (see: https://github.com/os-data/boost-paraguay/commit/a04e4f57c42f76890f1bd2ed673e0f1e780dffb9)
Multiple comments by @pudo and @pwalsh that using objects where possible provide benefits that simple arrays don't: https://github.com/openspending/fiscal-data-package/issues/93

# non-hierarchical dimension with one attribute mapping to several source fields
{
  "name": "function",
  "dimensionType": "classification",
  "attributes": [
    {
      "id": {
        "source": "function1_code"
      },
      "title": {
        "source": "function1_name"
      }
    }
  ]
}

# hierarchical dimension with several attributes mapping to several source fields
{
  "name": "function",
  "dimensionType": "classification",
  "attributes": [
    {
      "id": {
        "source": "function1_code"
      },
      "title": {
        "source": "function1_name"
      }
    },
    {
      "id": {
        "source": "function2_code"
      },
      "title": {
        "source": "function2_name"
      }
    }
  ]
}

If attributes is length == 1, we have a flat dimension object. if attributes is length > 1, we have an indexed hierarchy of dimension objects.

openspending / fiscal-data-package

Make dimension attributes an array of objects, each mapping several logical model fields to source CSV columns #97