Verify cumulative tests/fixtures are being normalized as expected

[damemi]

Noticed in #668 that cumulative normalization should drop the first data point it sees without a start timestamp. We should go through our other sum/cumulative fixtures and make sure the matching expect fixtures accurately represent a cumulative metric. If not, we should disable normalization or add additional points to get matching cumulatives in the output (except we are trying to test normalization dropping points)

[damemi]

One problem is that our test framework currently does one PushMetrics call, so I don't think we can test multiple CreateTimeSeriesRequests against the GCM api.

This means we basically can't test cumulative normalization this way, because if you try to create 2 data points in the same TimeSeriesRequest you will get a duplicate time series error.

The test framework could be updated to make multiple time series requests, which would be a cool enhancement.