Better monitoring for chunk state

[jml]

As part of understanding & fixing #61 and making sure it doesn't happen again, we want to gather data on:

• how big the backlog of flushes for chunks
• (maybe) how many chunks there are in the system
• how many chunks we're dropping

[juliusv]

(maybe) how many chunks there are in the system

That's just cortex_ingester_memory_chunks.

how many chunks we're dropping

That's rate(cortex_ingester_chunk_store_failures_total[1m]). During normal operation, failed stores are simply retried infinitely, but during shutdown, it's a one-shot thing.

[tomwilkie]

how big the backlog of flushes for chunks

Backlog can be inferred from the various metrics now.