dedup: don't inject a counter adjustment before any sample is emitted

[DeviousCardi]

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

Changes

newDedupSeriesIterator pre-seeds lastIter: a, useA: true before any sample has been emitted, while lastT is still math.MinInt64. Next() captures that as lastUseA and, if it then picks replica b, fires the counter-reset adjustment against replica a's first sample — a value that was never emitted. adjustAtValue sees a decrease, concludes "counter reset", and installs a permanent errAdjust equal to the inter-replica scrape offset.

The result is a counter flattened at the start of the window, so rate() under-reports. Query-frontend range splitting makes it visible in practice: each split gets a fresh iterator over a different first sample, and split starts are aligned to absolute interval multiples, so the dip lands at the same wall-clock boundaries every time.

This guards the adjustment on emittedAny := it.lastT != math.MinInt64, reusing the sentinel Next() already uses twice for the initial-penalty decision. Genuine mid-stream replica switches still adjust, so the #2401 mitigation is untouched.

Closes #9034

Verification

Synthetic counter increasing exactly 1/s across two replicas at 30s scrape with a 15s offset, so the true rate() is exactly 1.0:

	before	after
unsplit	1.000000	1.000000
split at 1800000	0.666667	1.000000
mismatching points	1 of 61	0 of 61

The corrupted sample is directly visible — t=1755000 v=1770.000 where it should be 1755.000, giving delta 30 over 45s.

Controls isolate it to this code path: a single replica, --deduplication.func=chain, and irate all return 1.000000 before and after.

New table case Regression test against 9034 in TestDedupSeriesSet, with replica b holding the earliest sample. Reverting only iter.go makes it fail with +3 on every sample — exactly 103 - 100, the inter-replica offset. TestDedupSeriesSet/Regression_test_against_2401 still passes.

go test ./pkg/dedup/... -count=1 and -race both pass; gofmt clean; no new go vet warnings.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 64.56%. Comparing base (050d688) to head (6f41f18).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #9041   +/-   ##
=======================================
  Coverage   64.55%   64.56%           
=======================================
  Files         289      289           
  Lines       37412    37413    +1     
=======================================
+ Hits        24153    24157    +4     
+ Misses      11158    11156    -2     
+ Partials     2101     2100    -1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.