Do not close and re-open TSDBs during ingester transfer

[pracucci]

When the ingester is leaving the ring and begins transferring TSDB blocks and WAL to a joining ingester, currently we do close the TSDB to flush it to disk before transferring. However, if a query hit the ingester after the TSDB has been closed and before the transfer is completed, the leaving ingester will re-open the TSDB (including WAL replay) which lead to an high resources consumption and query latency.

[pracucci]

I'm working on this along with @codesome.