Add support for a tunnel to expose inference server to remote calls

[iurisilvio]

Description

Embed a custom tunnelmole-client to expose a tunnel.

inference server start --tunnel --roboflow-api-key <api_key>

Our custom tunnelmole-service authorize the tunnel initialization based on roboflow api key.

Type of change

Experimental feature

• [x] New feature (non-breaking change which adds functionality)

How has this change been tested, please provide a testcase or example of how you tested the change?

I built from source and run it local. Works nice!

Easy to test calling some universe model with a remote image:

# inference will output some random url to be used here
export INFERENCE_HOST=https://somethingrandom.roboflow.run
curl -X POST "$INFERENCE_HOST/rock-paper-scissors-sxsw/14?image=https://source.roboflow.com/c8QoUtY71EUIn6gsXQMkSt8K0fC3/ST1zWYMcgt8HFqFBZNJy/original.jpg"

Any specific deployment considerations

I pull a custom tunnelmole image. This image is in roboflow GCR with public access.

Docs

Not changed docs yet, but new docs required.

• [ ] Docs updated? What were the changes:

[CLAassistant]

All committers have signed the CLA.

[iurisilvio]

I added some docs and also added a hook to stop the tunnel container when inference server stop.