Fix source generator evolver and hint-name failures

[erdtsieck]

What changed

• Treat a generated partial projection as covering its aggregate type so the snapshot-registration pipeline does not emit a second standalone evolver for the same aggregate.
• Reuse the aggregate constructor helper for generated Evolve/EvolveAsync fallback paths.
• Avoid emitting required-member object initializers for aggregate types without a public parameterless constructor.
• Normalize aggregate types discovered from snapshot/ref-aggregate call sites so nullable reference annotations do not leak into generated evolvers.
• Sanitize source-generator hint names before calling AddSource() so Roslyn-invalid characters such as ? cannot crash generation.
• Add regression tests for required-member primary-constructor aggregates with conventional projection Create/Apply methods and nullable [ReadAggregate] aggregate parameters discovered via IRefersToAggregate attributes.

Why

Two source-generator paths could break Marten 9 upgrade scenarios:

• A partial projection could generate the correct dispatcher while a later source-generator pipeline also emitted a fallback evolver for the same aggregate. That fallback did not know about the projection Create method and could generate construction code that fails for aggregates using required members and primary constructors.
• Aggregate types discovered from nullable parameters such as [ReadAggregate] EigenPrestatie? prestatie could feed ? into generated hint names, causing Roslyn to reject AddSource() and abort the whole generator pass.

Addresses JasperFx/marten#4542 and JasperFx/marten#4543.

Validation

• dotnet test .\src\JasperFx.Events.SourceGenerator.Tests\JasperFx.Events.SourceGenerator.Tests.csproj --no-restore

[jeremydmiller]

Verified this locally and it's the complete fix for marten#4542 (+ marten#4543). 👍

• Ran the suite on this branch: 18/18 green, including the reporter-faithful partial_projection_for_required_member_aggregate_suppresses_snapshot_fallback and the nullable_aggregate_parameter_attribute_does_not_crash_hint_name_generation (CS8785) case.
• I'd independently built a focused construction-only fix (Fix SG evolver construction for required-member aggregates (marten#4542) #362) with 3 extra regression tests (record→GetUninitializedObject, plain-class→default! initializer, reporter-shape compiles). This branch passes all 3 of those too → 21/21 when combined. So Fix source generator evolver and hint-name failures #359 is a strict superset; I've closed Fix SG evolver construction for required-member aggregates (marten#4542) #362 in favor of it.
• Also reproduced the underlying construction bug end-to-end against Marten release/9.0.0: a partial SingleStreamProjection over a required-member primary-ctor record fails to build (CS7036) with the released generator.

Two small things I'd double-check before merge (both look fine on read):

1. MarkSeen only suppresses the standalone evolver when Mode == PartialProjection && Methods.Count > 0 — so a Snapshot<T> with no covering partial projection still gets its evolver. ✅
2. The default! guard now keys off TypeKind.Struct || InstanceConstructors.Any(public parameterless).

Happy to push my 3 construction-focused tests onto this branch if you'd like the extra coverage.