Fix Hash Value For Prometheus Histogram Metrics.

[StevenYCChou]

Through internal investigation, I observed that one of the queue length is consistently high and blocked the whole queue manager. I identified that there are large volume of points with hash value 0 tried to be enqueued on to shard #0, which makes the sharding inbalance and the sharded queue are constantly blocked by shard #0.

When Stackdriver Prometheus Sidecar tries to transform Prometheus histogram metrics into Stackdriver timeseries with value type distribution, it should return the hash values derived by series cache rather than 0. This change fixes the undesired behavior.

[bmoyles0117]

This is great! Does this potentially apply to all of the time series that return a nil error? Specifically other calls to getResetAdjusted raise a potential concern in my mind.

[StevenYCChou]

This is great! Does this potentially apply to all of the time series that return a nil error? Specifically other calls to getResetAdjusted raise a potential concern in my mind.

My understanding is that if sampleBulder.next() returns both outSample as nil and error as nil, the caller PrometheusReader.Run() won't append the (Stackdriver) sample into queue. All returned value related to getResetAdjusted satisfies this condition.

[StevenYCChou]

Addressed the comments, please take another look. Thanks!