Bentoml: Add a command to containerize the model API server

[withsmilo]

Is your feature request related to a problem? Please describe.
BentoML supports various serving clusters now e.g. K8s, KNative, Kubeflow and KFServing. However we should execute some Docker commands to containerize our model API server. (Docs) How about adding a command to containerize our model API server to the BentoML instead of it?

Describe the solution you'd like

$ bentoml containerize [--docker-username ID] [--push] IrisClassifier:latest

Describe alternatives you've considered
N/A

Additional context
N/A

[parano]

That's a great idea @withsmilo! Really love it!

I guess with the sample command, you are also suggesting using the BentoService tag (e.g. IrisClassifier:20200518151822_B40DEC) as the docker image tag by default? That's really nice, it really helps users to keep the docker image tags (and especially the version) the same as the BentoService saved bundle's version. I'm not sure if this "name:version" string is always a valid docker image tag, we will need to check how we validate it and make sure the default versioneer function is generating valid docker tag string. Also, we can always add a parameter for user to set a docker image tag manually.

[withsmilo]

@parano
It's really nice idea, too!

[jackyzha0]

@parano would love to take this on too 👀