[java spring] Doesn't support different response models

Description

If you have a schema that has multiple responses with different models, the generated controller code only supports the first model.

eg. using the simple pet example (and changing the 'default' response to '400'):

      responses:
        '200':
          description: pet response
          schema:
            type: array
            items:
              $ref: '#/definitions/pet'
        '400':
          description: unexpected error
          schema:
            $ref: '#/definitions/errorModel'

Generates the following interface:

@ApiOperation(value = "", notes = "Returns all pets from the system that the user has access to", response = Pet.class, responseContainer = "List", tags={  })
    @ApiResponses(value = { 
        @ApiResponse(code = 200, message = "pet response", response = Pet.class),
        @ApiResponse(code = 400, message = "unexpected error", response = Pet.class) })
    @RequestMapping(value = "/pets",
        produces = { "application/json", "application/xml", "text/xml", "text/html" }, 
        consumes = { "application/json" },
        method = RequestMethod.GET)
    ResponseEntity<List<Pet>> findPets(@ApiParam(value = "tags to filter by") @RequestParam(value = "tags", required = false) List<String> tags,
        @ApiParam(value = "maximum number of results to return") @RequestParam(value = "limit", required = false) Integer limit);

Note that the error model isnt referenced anywhere in the generated code, and the response is set to ResponseEntity<List<Pet>> rather than the more generic Reponse class.

Swagger-codegen version

Latest head. Also the current version on http://editor.swagger.io/#/

Command line used for generation

generate -v -i ./swagger.yaml -l spring -o .

Start with the provided swagger.yaml example and run the documented generate -v -i ./swagger.yaml -l spring -o . command. Compare the generated Spring controller annotations and return type for the 200 and 400 responses; the completed behavior should represent both Pet and errorModel rather than using Pet for every response.