fix(gorilla-merger): serve cold queries after restart (empty-head MinTime + non-blocking reload)

[zzylol]

Summary

Cold parts were stored but a live cold query still returned empty after a merger restart, even with #341's MinTime floor in place and streamColdSeries debug logs added. Investigation on the live multinode cluster (node1) found TWO compounding root causes, neither addressed by #341:

1. customStore.timeRange() advertised math.MaxInt64 for an EMPTY tsdb head. tsdb.DB.StartTime() returns MaxInt64 when the head holds no samples (verified empirically) — exactly the state right after a restart, before the first warm fragment lands, while cold parts already exist in S3. That sentinel leaked into the StoreAPI's advertised MinTime, so thanos-query pruned the merger from every query (no window can be >= MaxInt64) and streamColdSeries never ran — the served-empty symptom. Fix: advertise the MIN of the tsdb StartTime and the cold-part floor, and never advertise MaxInt64 (fall back to MinInt64 when the store is genuinely empty so it stays discoverable and returns no series cheaply).

2. ColdPartStore.Reload blocked StoreAPI startup. It ran synchronously in main() before the gRPC/HTTP servers started, fetching + OpenPart-ing every stored part. With a large accumulated cold tier (live: ~2000 parts), this took ~53s during which the merger served nothing (warm or cold). Fix: run the reload in the background; the manifest is mutex-guarded so concurrent queries see a growing manifest until it completes, and the warm/open path serves immediately.

Live demonstration (node1)

	old image (#341)	fixed image
StoreAPI up after WAL replay	~53s (blocked on Reload)	~80ms
warm query during cold reload	EMPTY	returns data
cold-only-window query (post-reload)	n/a	series=1, streamColdSeries: emitting cold series cold_series=476

After restart with the fixed image: gorilla-merger up logged ~80ms after WAL replay while reloaded cold manifest (2014 parts) was still running in the background; warm count(http_requests_total) returned data immediately; a cold-only deep-window query_range returned the stored parts' series and the merger logged streamColdSeries: emitting cold series cold_series=476.

Test plan

• go build ./..., go vet ./..., gofmt -l clean
• go test ./... green
• New regression TestCustomStoreTimeRangeEmptyHead FAILS without the timeRange fix (empty head advertised MinTime = MaxInt64; merger would be pruned) and PASSES with it
• Live: cold-only query returns data + streamColdSeries invoked after a restart on the multinode cluster

🤖 Generated with Claude Code