Support endpoint subsetting

[ahg-g]

What would you like to be added:

Amend the endpoint picker protocol to define how a proxy can limit endpoint picking to a subset of pods.

Why is this needed:

The proxy may not be aligned with epp's view of what pods are ready and what are not. Only the intersection of their views is valid.

[ahg-g]

/assign @LiorLieberman

[hzxuzhonghu]

In this case, i want to supplement a basic requirement, sometimes gateway instances reside across zone/subzone, so it inference extension should be able to support locality aware routing too.

[robscott]

@hzxuzhonghu I think that would be a natural extension of this, but not necessarily something that needs to be codified.

For example, endpoint extensions could be deployed in each zone. Gateways could then use this protocol to provide zone-local endpoints in the subset that the endpoint extension could pick from. This kind of flexibility would allow the Gateway to provide a broader subset (including endpoints in other zones) if the local zone was overloaded.