This lack of availability is different from the availability in the A of CAP, since that availability holds only so long as the network is not partitioned (by definition in a CA system).
Such a system might not be considered a distributed system at all (although it may still be distributing load), since a partition-intolerant system is effectively one system as far as the CAP theorem is concerned.
So it's essentially a special case of the CAP theorem, but it is still useful to describe it as CA.
Therefore it's not a CA system, but a C system.
If there is some time period during which requests are not responded to within a time bound, the system is not available then, and further is not a 'highly' or 100% available system. That is what the CAP theorem is talking about.
Consistency, similarly, is not a state but a property that holds across all responses. Either you return a consistent response to all your requests, or you don't. In the context of CAP, there is no middle ground.