One Size Does Not Fit All:  Optimizing Sequence Length With Recurrent Neural Networks For Spectrum Sensing