Move custom attribute materialization and invocation to managed code

[jkoritzinsky]

Summary

Layer 2 of the reflection → custom attributes → func-eval → cleanup stack for #123864, building on am11's work in #126542 and the requested split in #126542 (comment).

• Move both CustomAttribute_CreateCustomAttributeInstance and CustomAttribute_CreatePropertyOrFieldData, including their exclusive blob/array decoding closure, into ordinary C#.
• Reuse the bounded managed blob reader without constructing intermediate custom-attribute data graphs. Preserve scalar widths, enum identity, typed arrays, named-argument ordering, exception boundaries, and type-resolution behavior.
• Keep fixed primitive/enum constructor arguments unboxed, with InlineArray16<T> storage, a bounded heap fallback, and explicit GC reporting for the byref vector. Box only where an object is required.
• Resolve type names through the existing span-based resolver, preserving contextual-reflection and collectible-type rules.
• Remove the obsolete materialization QCalls and temporary CA UCO bridge. Retain native CaArg/CaValue parsing and attribute-usage parsing.

Validation

Windows x64 Checked CoreCLR/Release libraries: live custom attributes, Reflection.Emit custom-attribute blobs, data-only reflection, and Reflection.Extensions coverage. Constructor/exception tests span the new 10,000-call promotion boundary. Coverage includes malformed/truncated blobs, enum widths, floating-point bits, moving GC, the 17-argument heap fallback, and collectible/contextual resolution. Forced shared and forced emitted invocation were also exercised.

Browser/WASI managed CoreLib and generated ABI inputs were checked; actual browser/WASI and big-endian runtime execution were not performed.

Performance evidence

Matched Release measurements compared managed decoding against a native-materialization baseline that already retained unboxed argument storage. On the local Windows x64 EPYC 7763 Hyper-V VM, warmed reference-argument, named-argument, array, and tagged-object cases measured approximately 0.73–0.82× the baseline time, with unchanged managed allocations. Fresh-process results were mixed and include parser/boxing-cache initialization; this is not a universal startup or cross-platform performance claim.

The next layers are #133659 (func-eval) and #133660 (dead native machinery cleanup).

---

_{Stack created with GitHub Stacks CLI • Give Feedback 💬}

Note

This PR description and implementation were prepared with GitHub Copilot. Commits that reuse am11's code retain the requested co-author attribution.

[azure-pipelines]

Azure Pipelines:
Successfully started running 3 pipeline(s).
13 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[dotnet-policy-service]

Tagging subscribers to this area: @steveisok, @dotnet/area-system-reflection
See info in area-owners.md if you want to be subscribed.