Add consecutive count check for Auto-scaling (#1703)

[sre-bot]

cherry-pick #1703 to release-1.1

---

What problem does this PR solve?

ref #1651

Noteice
As Auto-scaling is under alpha feature, the first target is to achieve basic auto-scaling Availability.
The extra control to avoid jitter for production-ready is not considered in this request. These could be recorded in this issue first.
#1651

What is changed and how does it work?

Add 3 new configuration for Auto-scaling API and default them them Controller.

1. MetricsTimeDuration which is used in the Prometheus Query
2. ScaleOutThreshold
3. ScaleInThreshold

ScaleOutThreshold/ScaleInThreshold is used to check the consecutive count in auto-scaling control to avoid jitter.

I also add the unit test for the current sync logic for tidb and tikv and the newly added function in util.go

In this request, each auto-scaling reconciliation would record the consecutive count in the target tidbclusterAutoScaler annotations marked as ScaleOutCount / ScaleInCount

For example:
In one tikv auto-scaling reconciliation,
if recommended replicas > current replicas, then:

tikvScaleOutCount = tikvScaleOutCount + 1;
tikvScaleInCount = 0;

if recommended replicas < current replicas, then:

tikvScaleOutCount = 0;
tikvScaleInCount= tikvScaleInCount +1;

if recommended replicas == current replicas, then:

tikvScaleOutCount = 0;
tikvScaleInCount= 0;

When the *ScaleOutCount/ *ScaleInCount is bigger than the defined property in autoscaler api, the auto-scaling would be performed and the record would be empty and recount.

Does this PR introduce a user-facing change?:

NONE