PgTargetTempTests: the temp-spill e2e is anchored at the planted window's end, so analyze_server measures the series that was planted rather than one that UtcNow slid past; and the observed-time pin names the one nominal-span use (the trend midpoint) instead of forbidding the expression (#3542 lane 6 follow-up to #3681)

[erikdarlingdata]

What does this PR do?

#3681 (lane 6 of #3542) was merged on its first commit while its own fix commit was still landing on the branch, so dev's Darling PostgreSQL tests now carry two red assertions in PgTargetTempTests — both in the TEST, neither in the product. This is that fix commit, cherry-picked onto dev (test file only, 13 lines).

1. The e2e's analyze_server call was un-anchored. Without as_of, the tool ends its window at UtcNow, a minute or two after the planted windowEnd, so the window's START slid past the first planted minutes — each a 120 MiB step in temp_bytes — and the frozen headline read 27.8 GB where the collector-alone assertions (which use the planted window) had just measured 28 GB. The call now passes as_of: windowEnd (the idiom lanes 2 and 3 already use), so the tool analyzes exactly the planted window. An anchored run is exploratory (not persisted); the test reads nothing persisted, and drill-down enrichment runs before the persistence gate, so the pg_temp_spill_statements assertion is unaffected.

2. A source pin over-matched. DoesNotContain("TimeRangeEnd - context.TimeRangeStart") was meant to pin "rates divide by observed time, never the nominal window" — but the coordinator-requested tps_trend legitimately uses the window's span ONCE, for the midpoint bound ($4) that splits the halves. The pin now says what it means: that expression appears exactly once, as (… ) / 2, and nothing divides by a TimeSpan of it (.TotalMilliseconds / .TotalSeconds / .TotalHours absent).

Which component(s) does this affect?

• Lite
• Darling
• Lite Tests
• Darling Tests
• SQL collection scripts
• Documentation
• Full Dashboard (deprecated)
• CLI Installer (deprecated)

How was this tested?

Test-only. Darling/Darling.Tests builds with zero warnings. The anchored shape was the one my local run of the whole pipeline against a throwaway PostgreSQL 18 store used (asOfUtc: windowEnd), which is why the pinned numbers matched there and not on the un-anchored CI run. First execution of the corrected assertions is this CI run.

Checklist

• I have read the contributing guide
• My code builds with zero warnings (dotnet build -c Debug)
• I have tested my changes against at least one SQL Server version (n/a — PostgreSQL-store test)
• I have not introduced any hardcoded credentials or server names