Translate ToRangeSet() from Int32Set, Int64Set, DateSet, LocalDateSet and YearMonthSet (split from #407) #487

Problem

ToRangeSet() on Int32Set, Int64Set, DateSet, LocalDateSet and YearMonthSet runs in memory only. #407 holds both bridge directions together with enumeration, but the two directions aren't alike:

• Multirange to set has to enumerate. Its input may be unbounded, and it yields one element per step.
• Set to multirange doesn't. Its input is finite, and its output is never larger.

Translating this direction also makes cross-family questions translate through operators that already run as SQL: b.Dates.ToRangeSet().IsContainedBy(b.Stay) ("are all the booked days inside the stay?") or b.Dates.ToRangeSet().Overlaps(closedPeriods).

Measured (PostgreSQL 16.15 and 18.1)

• (SELECT range_agg(int4range(x, x, '[]')) FROM unnest('{1,2,3,5}'::int[]) x) is {[1,4),[5,6)}.
• int.MaxValue fails: integer out of range, because the canonical form needs x + 1. long.MaxValue is the same case.
• The empty set gives NULL (range_agg over no rows), where memory gives RangeSet.Empty. It needs COALESCE(…, '{}').
• A NULL element in a row another tool wrote gives int4range(NULL, NULL, '[]'), which is (,). So {1,NULL} bridges to {(,)}, the whole domain. It must be filtered with WHERE x IS NOT NULL, as Translate value set Intersect, Except, Add and a canonical Union for the non-text families with composed SQL #478's canonical form does.
• DateOnly.MaxValue, which Npgsql writes as infinity, aggregates to [infinity,infinity]; 9999-12-31 itself gives [9999-12-31,10000-01-01). Both modes of Npgsql's infinity conversions need a test.
• YearMonthSet is stored as month-aligned dates, so its one-element range spans a month, not a day. Not measured yet.

Change

Translate as COALESCE((SELECT range_agg(<range>(vrN.x, vrN.x, '[]')) FROM unnest(set) AS vrN(x) WHERE vrN.x IS NOT NULL), '{}'), inside #478's NULL-over-NULL wrapper. Build it after EF Core's alias pass, as #478 builds its expressions, behind a placeholder that fails closed. At int.MaxValue and long.MaxValue, compare with what memory returns and what storing that result does (the domain-boundary table in docs/architecture.md). Then decide whether the server error is the documented answer.

Does it change an answer?

No. A predicate that throws today gets the in-memory answer.

Done when

• ToRangeSet() translates for all five set types, core and NodaTime
• a live test on PostgreSQL 16 and 18 agrees with memory, including the empty set, a NULL column, a NULL element written by another tool, and the extremes in both infinity modes
• docs/efcore.md#what-runs-where moves this direction out of the bridge row, and Translate enumeration and the bridge conversions — a batch after 10.3.0 #407 narrows to enumeration and the reverse bridge

Split out of #407.