Ensure ingester's TSDB Compact() is called after WAL replay

[pracucci]

When running the Cortex blocks storage, at ingester startup we do replay the WAL for any tenant's TSDB found in the local filesystem and, once done, the ingester switches to READY and start receiving samples from distributors.

There are scenarios where an ingester's TSDB may have a compactable HEAD right after the WAL replay but the HEAD is not compacted before the ingester switches to READY, because it just gets compacted after 1 minute as effect of the internal ticker used by compactionLoop().

I propose to ensure that the ingester compacts the HEAD (if compactable) after the WAL replay and *before switching to READY.