Stringize drops a node's Codomain, so domain(x, ZZ) prints as x and does not round trip #1022

Stringize prints no node's Codomain, so a node that carries one prints as though it did not, and
parsing the result gives a different node back. Same shape as #873, which was the Rational/Divf
case, and the last one I can find.

Measured on v2.3.0 (6b93b40):

var d = MathS.FromString("domain(x, ZZ)");
d.GetType()          // Entity.Variable
d.Codomain           // Integer
d.Stringize()        // "x"
MathS.FromString(d.Stringize()).Codomain   // Any
d == MathS.FromString(d.Stringize())       // False

The parser has the syntax already — domain(expr, SET) maps straight onto WithCodomain, and it
works for every node, not only variables:

domain(x + 1, ZZ)     -> Sumf,     Codomain = Integer
domain(sqrt(-1), RR)  -> Powf,     Codomain = Real
domain([1, 2], RR)    -> Matrix,   Codomain = Real

so this is only the printing half being absent.

It is not confined to a node somebody wrote domain(...) around, because the codomain travels with
the subnode:

var e = MathS.Sin(MathS.Var("x").WithCodomain(Domain.Integer)) + MathS.Var("y").WithCodomain(Domain.Real);
e.Stringize()   // "sin(x) + y" -- both annotations gone

Codomain is not decoration: it decides evaluation. sqrt(-1) is i, and the same expression with
Codomain = Real is NaN — the example in Domains.Definition.cs says exactly this, and prints the
same string for both.

Why it matters beyond the round trip: the printed form is the library's exchange format, so anything
that writes an expression out — a file, a database column, EntityJsonConverter (#323) — loses the
annotation silently, and a caller who narrowed a domain gets it back widened with no error anywhere.

StringizeRoundTripsToTheSameNode in EveryNodeSurvivesEveryPipelineTest does not catch it because
the samples it builds all carry the default codomain for their type. A node with a narrowed one is
worth adding there whichever way this is fixed.

Two ways to fix, and they are not equivalent:

1. print domain(<inner>, SET) wherever a node's codomain is not the one the parser would give the
   same text. Round trip becomes an identity; output for the ordinary case is unchanged, since the
   ordinary case has the default.
2. decide that Codomain is not part of a node's identity and take it out of record equality — which
   is roughly Unify Codomain with a "provided ... in RR" condition #721's direction, where the constraint becomes a provided ... in RR condition that
   is printed.

The first is the smaller fix and does not prejudge #721.