Statsdreceiver outputs cumulative temporality (incorrectly) for Sums w/o start timestamp

[jmacd]

Describe the bug
The statsdreceiver has these lines:

	nm.Sum().SetAggregationTemporality(pdata.MetricAggregationTemporalityDelta)
	if isMonotonicCounter {
		nm.Sum().SetAggregationTemporality(pdata.MetricAggregationTemporalityCumulative)
	}

	nm.Sum().SetIsMonotonic(true)

and

	dp.SetTimestamp(pdata.NewTimestampFromTime(timeNow))

What did you expect to see?
This receiver always outputs delta temporality, so:

	nm.Sum().SetAggregationTemporality(pdata.MetricAggregationTemporalityDelta)
	if isMonotonicCounter {
	    nm.Sum().SetIsMonotonic(true)
        }

and

	dp.SetStartTimestamp(pdata.NewTimestampFromTime(lastIIntervalTime))
	dp.SetTimestamp(pdata.NewTimestampFromTime(timeNow))

where lastIIntervalTime is computed from the containing StatsDParser.

What did you see instead?
This outputs cumulative temporality when it should output delta temporality.
This receiver should set start timestamp to the beginning of each report window.

Note also
The Summary points output for histogram and timing events also is missing its start timestamp,
however that field is underspecified in OpenTelemetry and while it may be OK to fill in the "beginning of each report window" as the start time, it is not how a Prometheus server would fill out that field. We end up in the case of "degenerate cumulative" which is a cumulative series that resets on every data point as a delta series would do.
This remark is meant to help understand a data model problem with Summary points, which will come to light as this issue is fixed. This issue is meant to apply only to the problem with Sum points.

[mwear]

I opened a PR that attempts to fix the counter portion of this ticket. Feel free to assign this issue to me, at least, responsibility for the counter part of it.

[jmacd]

Thank you @mwear! For the Summary portion, we have a solution here: open-telemetry/opentelemetry-specification#2002

The result is that the statsdreceiver behavior is off-spec. My plan is to insert a new exponential histogram to address that issue.