Agent DB Refactor

For variant neutron agents, they share the same database model to store
basic information, configurations, as well as dynamic workload data.
Neutron-server depends on it to apply correct operations such as producing
API response, scheduling, and HA. Also, how to maintain, transport those
data is a performance factor to neutron. We believe the db model needs a
refactor, base on some thinkings of the two scenarios.

First, currently there is no attribute to identify agent logical status,
such as initializing, ready, data sync, or error. Once agent is connected
to neutron-server, only admin state and liveness is checked on server side.
That means neutron-server may apply operations on agent blindly.

Second, agents send heartbeat message to transport those data. However,
carrying/parsing static data and writing them to db everytime is
inefficient. We'd like to see static and dynamic data is well decoupled
in agent db model. Then agent only needs to send static data once, and
subsequent heartbeats only carry dynamic data.

Please see spec for more details.