Add a metric to track number of ready pods

[ahg-g]

What would you like to be added:

A metric emitted by the EPP to track the number of ready pods from the EPP PoV.

Why is this needed:

The EPP doesn't own the status of the InferencePool object, so it can't keep track of that there. Having a metric to track the number of ready pods is important for operators to track rollouts.

[smarterclayton]

We probably should plan to break that metric down by attributes the pool is aware of (i.e. not O(pods), but O(groups of pods under the pool)). We don't have such a label, but I can already see ones coming up like heterogenity and assignment.

[JeffLuoo]

I have an idea to add a new metric for inference pool

inference_pool_per_pod_queue_size{name=<inference pool name>, pod=<name of pod under the pool>}

which tells the total number queue for each pod under the pool. It can provide:

1. number of pods related to the pool.
2. queue length for each pod, and see if our scheduling algorithm (or the priority) works as intended.

WDYT? @ahg-g @smarterclayton