Query: wait for initial endpoint discovery before becoming ready

[pedro-stanaka]

Problem

We observed a race condition where Thanos Query components were marking themselves as ready before discovering any endpoints. This created a timing gap that could lead to query reporting incomplete results:

• Query pods become ready immediately upon startup
• Endpoint discovery happens asynchronously in the background
• Queries arriving between readiness and endpoint discovery will report result from part of the endpoints

Solution:
This commit modifies the query readiness behavior to wait for the initial endpoint discovery to complete before marking the pod as ready. This ensures that when a query pod reports ready, it has already attempted to discover and connect to available endpoints.

Impact:

• Positive: Eliminates the race condition where queries could be routed to Query pods that haven't discovered any endpoints yet
• Negative: Slightly increases startup time as pods won't be ready until endpoint discovery completes (typically up to a few seconds in normal conditions)

• I added CHANGELOG entry for this change.
• Change is not relevant to the end user.

Verification

New tests.

Relates to #8332

[pedro-stanaka]

Failed test seems a flake.

[pedro-stanaka]

We were testing this in our staging (and then production) env, and we found out we need to also trigger a DNS update before we start the store resolution, otherwise it is pointless. We should invest some effort into refactoring these things to share a single update loop, since they are so closely related.

Added a new commit to fix that, this represents what we have now in production, and so far it has been working well for us.

[GiedriusS]

Should we do the same with the resolver that is created in dns.RegisterGRPCResolver(...)? Basically, call r.resolve() and do all the things (move all that

func() {
			if err := r.resolve(); err != nil {
			...
}

Into a separate function and call that inside of Build()?