Change to metrics to keep track of when the metric value was first set

This CL also adds IncrementableMetric#reset() methods to allow resetting the
value and start timestamp of IncrementableMetrics.

This is necessary because some backends, like Stackdriver, use non-monotonic
changes in cumulative metric values to detect timeseries restarts. Tracking and
re-setting the start timestamp allows users to track mostly monotonic metrics
which may have non-monotonic discontinuities.

See https://cloud.google.com/monitoring/api/ref_v3/rest/v3/TimeSeries#Point for
more details.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130795229

java/google/registry/monitoring/metrics/VirtualMetric.java

@@ -29,6 +29,10 @@ import org.joda.time.Instant;
  *
  * <p>This pattern works well for gauge-like metrics, such as CPU usage, memory usage, and file
  * descriptor counts.
+ *
+ * <p>The {@link MetricPoint#interval()} of values of instances of this metric will always have a
+ * start time equal to the end time, since the metric value represents a point-in-time snapshot with
+ * no relationship to prior values.
  */
 @ThreadSafe
 public final class VirtualMetric<V> extends AbstractMetric<V> {
@@ -77,7 +81,8 @@ public final class VirtualMetric<V> extends AbstractMetric<V> {
 
     ImmutableList.Builder<MetricPoint<V>> metricPoints = ImmutableList.builder();
     for (Entry<ImmutableList<String>, V> entry : values.entrySet()) {
-      metricPoints.add(MetricPoint.create(this, entry.getKey(), timestamp, entry.getValue()));
+      metricPoints.add(
+          MetricPoint.create(this, entry.getKey(), timestamp, timestamp, entry.getValue()));
     }
 
     cardinality = values.size();