Fix: Handle dynamic batch sizes in TensorRT ReID model

[usamaimdadsian]

Previously, in the inference of TensorRT models if input batch exceeds the batch shape used for TensorRT conversion it resulted in error on the next steps. My fix divides the input batch in sub batches and those sub batches are used by the model for inference. After the inference I stack the features of ReID models and return them.

[mikel-brostrom]

Great! Thanks for contributing @usamaimdadsian 🚀