Generalize check for dynamic scopes when saving EH clauses

[jakobbotsch]

The Task returning thunks generated for runtime async methods have a catch (Exception ex) in them. During EH we were lazily trying to resolve the dynamic type token to a type handle using the metadata scope from the original IL method. Ensure we do this during JIT time instead, using the right resolver.

Fix #115779

[dotnet-policy-service]

Tagging subscribers to this area: @mangod9
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull Request Overview

This PR generalizes exception handling (EH) clause token resolution by switching from a method-specific dynamic check to a scope-based approach, ensuring the correct resolver is used at JIT time.

• Replace m_pMethodBeingCompiled->IsDynamicMethod() with IsDynamicScope(m_MethodInfo.scope)
• Switch from AsDynamicMethodDesc()->GetResolver() to GetDynamicResolver(m_MethodInfo.scope)
• Resolve dynamic type tokens during JIT compilation rather than lazily at EH time

Comments suppressed due to low confidence (2)

src/coreclr/vm/jitinterface.cpp:12780

• Add tests that cover EH clause resolution for both dynamic and non-dynamic scopes to verify ResolveToken is invoked only when intended.

    if (IsDynamicScope(m_MethodInfo.scope) &&

src/coreclr/vm/jitinterface.cpp:12785

• Potential null dereference: GetDynamicResolver may return null if the scope isn't dynamic. Consider adding a null check or assertion before calling ResolveToken.

        GetDynamicResolver(m_MethodInfo.scope)->ResolveToken(clause->ClassToken, &resolved);

[VSadov]

I have a testcase for the repro in https://github.com/VSadov/runtime/tree/asyncEhBug. May be useful to cherry-pick into this PR.

[VSadov]

The test is passing on my machine now (with this change).

[VSadov]

LGTM. Thanks!

[jakobbotsch]

I have a testcase for the repro in https://github.com/VSadov/runtime/tree/asyncEhBug. May be useful to cherry-pick into this PR.

I have not generally been adding regression tests for these very basic issues found by Fuzzlyn. Fuzzlyn finds these issues in a few seconds, so I think it would be better to get the coverage by adding Fuzzlyn testing of runtime async to the CI (it's on my TODO list...).