Added -store.max-chunks-per-query limit support to blocks storage

[pracucci]

What this PR does:
In this PR:

• Added -store.max-chunks-per-query limit support to blocks storage
• Removed max-sample-count support because it's flawed in Thanos (it's per block and not per query as documented). I will work on this in a subsequent PR (I need to fix it in Thanos first) and the limit in store-gateway will be based on -store.max-chunks-per-query too.

Out of the scope of this PR:

• Add the limit to the store-gateway too (will be done in a separate PR, because I need to refactor and fix Thanos first)
• Add a negative cache to avoid running the same query for the same tenant if we already know it will hit the max chunks limit (issue)

Which issue(s) this PR fixes:
Fixes #2844

Checklist

• Tests updated
• Documentation added
• CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX]

[pstibrany]

LGTM, good job.

(I'm becoming more nervous about mixing direct and atomic access to the same variable, although in this specific case, there is clear happens-before relationship between the two, so I think it is correct. [Talking about numChunks in fetchSeriesFromStores method])

[pracucci]

LGTM, good job.

Thanks!

(I'm becoming more nervous about mixing direct and atomic access to the same variable, although in this specific case, there is clear happens-before relationship between the two, so I think it is correct. [Talking about numChunks in fetchSeriesFromStores method])

It's safe because of the g.Wait() before reading numChunks, but let's be consistent: I've committed the usage of atomic.LoadInt32() when reading it in the return too.

[pstibrany]

(I'm becoming more nervous about mixing direct and atomic access to the same variable, although in this specific case, there is clear happens-before relationship between the two, so I think it is correct. [Talking about numChunks in fetchSeriesFromStores method])

To expand on this comment, I think my reasoning is wrong. Program happens-before order, and atomic accesses are independent, so correct version of the code should use atomic load at the end of fetchSeriesFromStores. It would be cleaner to use uber-go/atomic package instead for this.

But no need to change it in this PR... we have similar code elsewhere.

[pstibrany]

It's safe because of the g.Wait() before reading numChunks

This is what I thought as well. Unfortunately, there is no such guarantee.