Migrate scalar builtin types to generated TypeDef assembly format ownership

[Copilot]

Scalar builtin types (float, index, none) had split ownership: generated TypeDef classes existed and were registered, but binding still went through hand-written factory delegates and FloatTypeReference held a catch-all TypeDefinition = new("float"). This completes the migration so generated TypeDef definitions own scalar binding end-to-end via assembly formats, matching the pattern already established by IntegerType.

New assembly format classes

• BuiltinIndexTypeAssemblyFormat — binds BuiltinIndexTypeSyntax → IndexType
• BuiltinNoneTypeAssemblyFormat — binds BuiltinNoneTypeSyntax → NoneType
• BuiltinScalarFloatTypeAssemblyFormat — shared format parameterized by a constructor delegate; handles all 18 float variants without per-type subclasses

Metadata-driven generation

BuiltinTypesExtensions.td now supplies csharpAssemblyFormat for all scalar float types plus Builtin_Index and Builtin_None. The generator (TypeEmitter) uses this metadata to:

• emit assembly format instead of factory delegate
• fix float constructors to use the MLIR mnemonic ("f32") rather than the qualified registry key ("builtin.f32")

// Before: factory-backed, wrong name in constructor
new TypeDefinition("builtin.f32", factory: static context => Float32Type.BindValue(context));
public Float32Type(...) : base(TypeDefinition.Name, ...)  // "builtin.f32" — wrong

// After: assembly-format-backed, correct mnemonic
new TypeDefinition("builtin.f32", new BuiltinScalarFloatTypeAssemblyFormat(syntax => new Float32Type(syntax)));
public Float32Type(...) : base("f32", ...)  // correct

Binder registry lookup

Binder.GetStructuredTypeDefinitionName now maps scalar syntax to "builtin.integer", "builtin.f32", "builtin.index", "builtin.none", etc., so the registry lookup resolves correctly to the generated builtin TypeDefinition.

FloatTypeReference cleanup

Removed the catch-all TypeDefinition = new("float") static property. The base class now returns null for Definition; generated subclasses override with their own TypeDefinition. Fallback-bound FloatTypeReference values (no registry) thus have Definition == null.

Test coverage

• Updated existing registered-binding test to use the correct "builtin.integer" lookup key
• Added tests for: registered float/index/none → generated typed subclasses with non-null Definition; fallback float → Definition == null; all generated scalar TypeDefinitions have AssemblyFormat != null; registered and fallback scalar types compare equal

[github-actions]

TableGen Interpreter Benchmarks

Commit: cc5dd2ea298b9679bfe3ea06f107bc85840e4bdb

Benchmark	Baseline (ms)	Candidate (ms)	Delta	Status
Evaluate.PathologicalDeepClassChain	0.028	0.028	-0.1%	Flat
Evaluate.PathologicalListPipeline	0.158	0.152	-3.4%	Flat
Evaluate.PreludeArithOps	9.358	8.825	-5.7%	Improvement
Evaluate.PreludeMiniDialect	3.463	3.501	+1.1%	Flat
Evaluate.SimpleInheritanceLets	0.002	0.002	+6.1%	Regression
Evaluate.UtilsTd	0.019	0.018	-1.2%	Flat
ParseAndEvaluate.PreludeArithOps	21.629	22.328	+3.2%	Flat
ParseAndEvaluate.PreludeMiniDialect	9.289	9.287	0.0%	Flat

Interpretation:

• Regression means candidate mean time is more than 5% slower than baseline.
• Improvement means candidate mean time is more than 5% faster than baseline.
• Flat means the change stayed within a 5% noise band.