[CaffeinatedCoder]

Closes #451

Stacked on #455 (#450): both edit claim E76, on adjacent lines. Merge #455 first.

Problem

1. CalendarFactoryParityTests had one seed, and it proves only the extremes half. Nothing proved the refusal half, CalendarFactories_OverAnArgumentMemoryRefuses_RaiseInPostgres. That half is the one most likely to go vacuous: it accepts three SQLSTATEs, and its Where form passes if any row raises.
2. docs/efcore.md#what-runs-where says "filter, join and sort by it", but no test covered a join.
3. The existing seed repeated its diff --git line.

Change

• New seed test/oracles/seeds/calendar-iso-week-53-of-a-52-week-year-rolls-over.patch (Oracles: CalendarFactoryParityTests). It changes week <= weeks to week <= weeks + 1 in the ISO week check, so IsoWeek(2025, 53) rolls over to 2026-W01.
• New test CalendarFactories_AsAJoinKey_PairTheRowsMemoryPairs. It joins Month(x.Day) with Month(y.Year, y.Month), and does the same for Quarter and Year. The pairs must equal memory's, and the rows at DateOnly.MinValue and MaxValue must find a partner. E76's claim now names the join.
• The existing seed has one header line. Its Fails: line now names the join, which it also fails.

Tests

On PostgreSQL 16, with main's RC2 SDK:

• The new seed fails CalendarFactories_OverAnArgumentMemoryRefuses_RaiseInPostgres and nothing else (1 of 4).
• The old seed fails InEveryQueryForm and the new join test (2 of 4); the join gives 42 pairs where memory gives 67.
• dotnet run tools/oracles.cs seeds calendar reports every seed is caught.
• Without a seed, the class passes (4 of 4), and so do the conventions tests (417).

🤖 Generated with Claude Code