Tiny-PULP-Dronets : Squeezing Neural Networks for Faster and Lighter Inference on Multi Tasking Autonomous Nano-Drones