Is your feature request related to a problem? Please describe.

Currently, the BentoML API model server does not print the errors and stack trace when the exception was raised within the user's inference API function code. This makes it hard for users to debug issues in their code.

**D

Deploy kfserving v0.3.0 and sample app

kubectl apply -f kfserving.v0.3.0.yaml
kubectl apply -f sklearn.yaml

I have k8s cluster with enabled PSP and ImagePullAlways plugin
https://kubernetes.io/docs/reference/access-authn-authz/admission-controllers/
After all kservice isn't working

kubectl describe replicaset.apps/sklearn-iris-predictor-default-x9v87-deployment-784dbd

@zacbrannelly Not sure if this is possible in the current released version, so creating this just to keep track.