Add auto-scaling calculation based by CPU load (#1722)

[sre-bot]

cherry-pick #1722 to release-1.1

---

What problem does this PR solve?

#1651
Adding auto-scaling calculation algorithm based by CPU load.

This request adds the realization of auto-scaling calculation algorithm based by CPU load.
Currently, the auto-scaling is designed to support the cpu AverageUtilization in the following api spec:

  tikv:
    MetricsTimeDuration: "5m"
    metrics:
    - type: "Resource"
      resource:
        name: "cpu"
        target:
          type: "Utilization"
          averageUtilization: 80

notice
To use the auto-scaling by CPU load, the CPU requests in resources must be defined.

What is changed and how does it work?

I would take this spec as an example to explain how the calculation work.
The users create it to claim that the Operator could auto-scale tikv by CPU average load and the threshold value of average Utilization is 80%.

The Operator would query the CPU metrics from Prometheus (currently) to calculate the sum of CPU seconds for each tikv instances in the past 5mins every resync duration seconds.
If the sum of CPU seconds for total tikv instances in the past 5mins is 500 sec as cpuSecsTotal The cpu requests in resources is defined as 1 (1000mi) and current replicas of tikv is 3.

cpuRequestsRatio = 1000 / 1000 =1
durationSecs = 5mins = 300s
utilizationRatio = 80%
expectedCpuSecsTotal = cpuRequestsRatio * durationSeconds * currentReplicas * utilizationRatio = 1 * 300 * 3 * 0.8 = 720
recommendedReplicas = ceil(cpuSecsTotal * currentReplicas / expectedCpuSecsTotal) = 3

Related changes

• Need to cherry-pick to the release branch

Does this PR introduce a user-facing change?:

NONE