This simplification changes bindings so they all change from looking like this:

...
"<node/edge/path>_bindings": {
          "key1": [
            {
              "id": "<CURIE/EdgeID/AuxGraphID>"
            }
          ],
          "key2": [
            {
              "id": "<CURIE/EdgeID/AuxGraphID>"
            }
          ]
        },
...

to this:

...
"<node/edge/path>_bindings": {
          "key1": {
            "ids": ["<CURIE/EdgeID/AuxGraphID>"]
          }
          "key2": {
            "ids": ["<CURIE/EdgeID/AuxGraphID>"]
          }
        },
...

The reasoning for keeping the binding objects, rather than making it a simple id: array, is to leave room for possible future changes to set_interpretation response formatting. A likely-enough future change would be to use unified binding objects so attributes can be attached that reference the set of nodes/aggregation of edges (instead of the current format of using reasoner-constructed Nodes and Edges with support graphs). This format allows for that possibility, as well as testing prior to future TRAPI changes because the binding objects allow for arbitrary additional properties.