Previously needsDedup() could return true if the de-duplication wasn't needed for the following case:
d < interval
/ \
| v | v |
interval interval
Now it properly returns false for this case
Now it leaves only the first data point on each `-dedup.minScrapeInterval` interval.
Previously it may leave two data points on the interval. This could lead to unexpected results
for `histogram_quantile(phi, sum(rate(buckets)) by (le))` query.
