[proposal] resource recommendation controller for predicting resource requirement for K8s workloads.

[zwzhang0107]

What is your proposal:
Collecting history metrics to estimate resource usage of containers or tasks, which can help the right-sizing and scheduling optimization for load-balancing.

Why is this needed:
#1880

Is there a suggested solution, if so, please add it:
Development stages:

• proposal of metric prediction framework
• define MetricPredtion API
• implement of MetricPredtion framework controller with cpu and memory prediction of workload in container level, based on metric server source.
• implement of Recommendation controller which supports fine-grained strategies during MetricPredcition for different kinds of workload in K8s, such as workload type(Job/Service), OOM event.