store: Adapt for objects changes (removals) faster

[bwplotka]

Basically we sync metas every 3 minutes, but it can happen that compactor or repair command remove some unnecessary block in the time between syncs. This result in temporary query error: receive series: rpc error: code = Aborted desc = fetch series for block 01C8DBKCT8TVC7KRDBC9TWBTPJ: preload postings: read postings range: get range reader: storage: object doesn't exist

How to improve that?

• decrease the interval
• catch not found obj error or check for existence of block and handle that somehow (ignore these postings?)

[stevenkehrli-trelnex]

We're seeing this with master-2018-06-15-5b66b72.
Will it eventually resolve itself? Message has repeated every 5 minutes for over 30 minutes now.

[bwplotka]

We need this to solve this issue: #700

[bwplotka]

Closing in favor of: #700