[FocoosModel][Processor] support batch inference

[CuriousDolphin]

call method of FocoosModel should return a list of ModelOutput (or np_array) instead of FocoosDetection, in this way we can easly support batch inference.

related to #132 #104