Separate document corpus definition from indices

[danielmitterdorfer]

Currently the data file definition is tied to an index type. E.g.:

  "indices": [
    {
      "name": "geonames",
      "types": [
        {
          "name": "type",
          "mapping": "mappings.json",
          "documents": "documents.json.bz2",
          "document-count": 11396505,
          "compressed-bytes": 264698741,
          "uncompressed-bytes": 3547614383
        }
      ]
    }
  ]

To allow for more flexibility we should separate them, e.g. by defining them in a separate corpora block. This would also give us more flexibility in allowing other formats (e.g. snapshots, see #341):

"corpora": [
  {
    "name": "geonames",
    "documents": [
      {
        "source-file": "documents.json.bz2",
        "document-count": 11396505,
          "compressed-bytes": 264698741,
          "uncompressed-bytes": 3547614383
      }
    ]
  }
]

We will also add an additional parameter to the bulk index runner to tell it which document corpus it should use (called corpora).

Note: For backwards-compatibility we should not remove the old document-related properties from our standard tracks yet.

[danielmitterdorfer]

Note: Ensure that scenario described in #325 is supported by our implementation.