Multithreaded Python API and Pylot Example

[haixuanTao]

This PR adds:

• a couple of examples, with long running CPU/GPU task such as an object detection function.
• transform the previous single-threaded server into a multithread server.
• Add a bench to see the time taken by the Python binding as we have seen some huge time sink within it.
• Refactor some of the previous code,
• Make a Docker to test Python example

To run the docker container

# Within dora root folder
nvidia-docker build --tag dora .
nvidia-docker run -itd --name dora -p 20022:22 dora /bin/bash
nvidia-docker exec -itd dora /home/erdos/workspace/pylot/scripts/run_simulator.sh
nvidia-docker cp ~/.ssh/id_rsa.pub dora:/home/erdos/.ssh/authorized_keys
nvidia-docker exec -i -t dora sudo chown erdos /home/erdos/.ssh/authorized_keys
nvidia-docker exec -i -t dora sudo service ssh start
ssh -p 20022 -X erdos@localhost

• Within the docker container, to run the object detection simulation that use ~6Gib of GPU memory:

cd /home/erdos/workspace/dora-rs
export PYTHONPATH=$PYTHONPATH:$(pwd)/examples/pylot
cargo run start-python carla_source_operator send &
cargo run start-python efficient_detection run image &
cargo run start-python sink_eval_plot plot destination

• Within the docker container, to run just a visualisation of the simulation:

cd /home/erdos/workspace/dora-rs
export PYTHONPATH=$PYTHONPATH:$(pwd)/examples/pylot
cargo run start-python carla_source_operator send &
SOURCE=image cargo run start-python sink_eval_plot plot image

[phil-opp]

Looks good to me!