Enhanced exception structure and handling

Presently, there is very little structure in the exceptions used in kuryr-kubernetes. It would be helpful to have a unified structure to all exceptions that are expected to be raised as part of normal kuryr operation. This document proposes a structure with severity levels and logical inheritance to better reflect the cause of and action taken for exceptions.

The proposed class structure can be seen at the spec details url.

- isFatal controls if the exception will cause kuryr-controller/kuryr-cni to crash or enter unhealthy state.
- RecoverableError can be fatal or not, depending on value returned by isRecoveryPossible() (which could internally implement a simple counting mechanism upto a threshold).
- This is useful for temporary errors like unreachable api's (k8s, keystone, neutron, ovs), race conditions, etc.
- Each exception under RecoverableError can implement its own isRecoveryPossible() method to have independent mechanisms.