add option to use the new chunk disk mapper from upstream

[replay]

This removes the OldChunkDiskMapper which we added a few months ago.

At the time we added it because we needed a way to disable the chunk write queue and in upstream this wasn't possible. In the meantime upstream has been modified in such a way that the chunk write queue can also be disabled there by passing a size of 0, so we should make mimir-prometheus match upstream again.

Edit:

I updated this PR, the current description is now this one: #216 (comment)

[pracucci]

I prefer to keep the OldChunkDiskMapper until we've migrated our prod cells to the new one (progressively). Reason why OldChunkDiskMapper back in the days (even if it was already supported to disable the queue-based writer with a value 0) was to have a safer way to progressively rollout to the queue-based one

[jesusvazquez]

When I brought prometheus/main changes into mimir-prometheus I had to add this to a test so that the behaviour would be the same

mimir-prometheus/tsdb/head_test.go

Line 1621 in 51f26e0

opts.ChunkWriteQueueSize = 1 // We need to set this option so that we use the async queue. Upstream prometheus uses the queue directly.

That would probably conflict in future merges so lets not forget to leave it the same as in prometheus/main once we're done with this PR.

[replay]

I prefer to keep the OldChunkDiskMapper until we've migrated our prod cells to the new one (progressively). Reason why OldChunkDiskMapper back in the days (even if it was already supported to disable the queue-based writer with a value 0) was to have a safer way to progressively rollout to the queue-based one

I updated this PR, now the only thing which it does is to add an option to use the new chunk disk mapper.
Note that this new option is independent of the chunk write queue size, that way we can first (slowly) roll out the new chunk disk mapper without activating the queue, if this works fine then we can remove the old chunk disk mapper. The primary goal of this is to make mimir-prometheus as similar to upstream as possible, the goal is not to roll out the chunk write queue again. In a later step we can consider re-activating the chunk write queue.
The default value of the new option is to continue using the old chunk disk mapper.

[pstibrany]

LGTM (no need to block on my comment)