Provide another abstract layer for Senlin driver for differentiation hiding

Add another abstract layer which hides interface differentiation between multiple drivers of the same type and provides unified interface for profile, e.g. alarm interfaces for scaling policy which can be mapped to both Ceilometer or Monasca driver; loadblancer interfaces for lb policy which can be mapped to both Neutron LBaaS or AWS LBaaS driver.

The basic idea is to map a request of common driver interface to a series of operations of a specific driver backend.

e.g. lb_create using `neutron_v2` driver type will be mapped to the following operations:
    1) creating loadbalancer:
       - neutron_v2.NeutronClient.loadbalancer_create()
    2) creating listener:
       - neutron_v2.NeutronClient.listener_create()
    3) creating pool:
       - neutron_v2.NeutronClient.pool_create()
    4) creating healthmonitor(optional):
       - neutron_v2.NeutronClient.healthmonitor_create()

We need to figure out a model to describe the mapping logic, including the list of functions need to invoke, the parameters list, and the sequence of function call.

OR we can just use Heat to implement this function?
 - Provide a mapping between device type and Heat resource_type;
 - Provide a operation mapping;

Not sure whether we should do this job in this way.