[BUG] ResponseTimeCircuitBreaker#slidingCounter is actually a 'rollingCounter' and cause unexpected degrade.

`public ResponseTimeCircuitBreaker(DegradeRule rule) {
this(rule, new SlowRequestLeapArray(1, rule.getStatIntervalMs()));
}`

The sampleCount of SlowRequestLeapArray is set to 1, this means the count number will be reset to **zero** when a new statistic window begins and makes it not really a **sliding** counter.

The problem is, when I mean to degrade when 100% of RTs exceed 5ms in 1 seconds( minRequestAmount is set to 1 ), in this case only one slow request can cause the resource degrade. Though a larger minRequestAmount mitigates this issue, but the proper number is not easy to estimate 'cause the QPS is changing.

So I suggest to set the sampleCount to at least 2.

`public ResponseTimeCircuitBreaker(DegradeRule rule) {
this(rule, new SlowRequestLeapArray(**2,** rule.getStatIntervalMs()));
}`

Start with the ResponseTimeCircuitBreaker constructor and its SlowRequestLeapArray configuration. Check how the one-second statistic window resets the count, then verify the proposed sample-count behavior with a focused regression test; done means slow requests are evaluated across sliding windows without an unintended reset to zero.

Written by the indexing model from the issue text.