Better handling for near duplicate influx metrics

[alexgreenbank]

This influx blog post on duplicate metrics has one suggestion of incrementing the timestamp by 1ns to ensure uniqueness.

Given that timestamps in Prometheus/Mimir have 1ms precision it is very unlikely (1 in 1,000,000) that an increase of 1ns will lead to a unique timestamp when it is translated/truncated to ms.

The proxy should check for timestamps that Influx would consider unique and do its best to ensure uniqueness in Mimir in a similar way (by incrementing timestamps by 1ms). Note this should only really be applied to near-duplicates it finds in an individual batch, not actual duplicates. Also inter-batch checks are beyond the scope of this initial work.