tests: a second seed for the calendar factories' refusals, and a join case for E76 #451

Found by a review of #191 (main PR #447 @ 879c82a, release/10.x PR #449 @ 684216b), checked on PostgreSQL 16.15 and 18.1.

Problem

1. The one seed proves only the extremes half of CalendarFactoryParityTests. calendar-last-period-ends-on-a-day-not-infinity.patch breaks the last period's constant. Nothing proves the refusal half, CalendarFactories_OverAnArgumentMemoryRefuses_RaiseInPostgres, which is the half most likely to go vacuous: it accepts three SQLSTATEs, and its Where form passes if any row raises. A second seed in MondayOf (ValueRangesMethodCallTranslator.cs) shows the class is real and is caught by that test alone:

   - sqlExpressionFactory.LessThanOrEqual(week, weeks)
   + sqlExpressionFactory.LessThanOrEqual(week, sqlExpressionFactory.Add(weeks, Int(1)))

   With it, week 53 of a 52-week year rolls over to the next year's week 1. On main @ 7503c18, PostgreSQL 16:

   Failed CalendarFactories_OverAnArgumentMemoryRefuses_RaiseInPostgres
   IsoWeek(2025, 53) Select: the server answered, where memory throws
   IsoWeek(2025, 53) Where: the server answered, where memory throws
   Failed: 1, Passed: 2
   

   The class's other two tests pass with the seed.

2. "Filter, join and sort by it" isn't held for the join. docs/efcore.md#what-runs-where claims a join (quoted by E76), but neither E76's claim nor any test covers one. It works: a join on DateRange.Month(x.Day) == DateRange.Month(y.Y, m) over both extremes gives the 1,365 rows memory gives. So this is a gap, not a bug.

3. Cosmetic: the seed patch repeats its diff --git header line, the only seed in test/oracles/seeds/ that does. git apply accepts it.

Change

• Add the seed above as test/oracles/seeds/calendar-iso-week-53-of-a-52-week-year-rolls-over.patch, with Oracles: CalendarFactoryParityTests. Then run dotnet run tools/oracles.cs seeds calendar on both lines.
• Add a join case to CalendarFactoryParityTests that compares with memory at both extremes, and add "a join" to E76's claim. Or drop "join" from the doc.
• Remove the duplicate header line from the existing seed.

Test

The new seed fails CalendarFactories_OverAnArgumentMemoryRefuses_RaiseInPostgres and nothing else. That is the proof verify-the-guard asks for.