feat: customize pod informer to reduce memory usage

[JackZxj]

Use a custom Pod Informer to limit the number of concurrent list pods from member clusters, and simplify the metadata of the pods before we store them into the cache.

The reference conversations: #14

As result:

4 clusters, 100 kwok-node/cluster, 10 deployment/cluster, 1000 pod/deployment, total: 10k pod/cluster

• case1: simplify the pod (In the last three rounds, kwok failed to delete pods of a cluster so the pods of cluster came to 20k; and some kwok nodes got not ready, so the available cpu,gpu came to zero, but the pods still in work. The overall impact is not significant.)
• case2: regular pod informer
• case3: simplify the pod and limit 2 available listers
• case4: simplify the pod, ignore modified event of watch, and limit 2 available listers

summary: It can save at least half of the memory, and reduce the rate of change of memory usage

[CLAassistant]

All committers have signed the CLA.

[gary-lgy]

The client-go informers use context.TODO(). Intuitively I think it does make sense to shutdown the listwatch when the federatedclient is shutting down, but I can't be sure. @JackZxj @limhawjia @SOF3 @mrlihanbo WDYT?

[SOF3]

I think we can reorganize all context usages in a separate PR; the scope is too big and doesn't need to be fixed here immediately.

[gary-lgy]

Agree that context usage should be fixed in a separate PR. I'm just wondering whether using ctx instead of context.TODO here is correct. Using ctx would mean the ListWatch would shutdown when ctx is cancelled.

[gary-lgy]

@JackZxj Great enhancement. Thanks for the contribution!