Introduce quorum-based chunk deduplication strategy

[hczhu-db]

This chunk dedup strategy is specific to Databricks' setup where each time series is written to at least 2 out of 3 replicas. Each chunk should have 3 replicas in most cases, and 2 replicas in the worst acceptable cases. Quorum-based deduplication is used to pick the majority value among 3 replicas. If a chunck has only 2 identical replicas, there might be another chunk with corrupt data. We want to send those two identical replicas to the later quorum-based deduplication process to dominate any corrupt third replica.

Unit test

$ go test -timeout 30s -run ^TestDedupRespHeap_QuorumChunkDedup$ github.com/thanos-io/thanos/pkg/store

ok  	github.com/thanos-io/thanos/pkg/store	2.364s

Integration test with the new feature

[jnyi]

great job HC, awesome debugging!

[hczhu-db]

Addressed the comments