Narrow Mono AOT typeload error cleanup

[vitek-karas]

Summary

• consume the expected recoverable MissingField / BadImage metadata errors after lowering Mono AOT field-resolution failures into runtime throw IR
• add a direct Mono mini regression in src/mono/mono/mini/iltests.il that covers a generic missing-field access followed by a later accepted inline in the same compiled method

Testing

• ./build.sh mono -os iossimulator -arch arm64 -c Release -rf mono
• validated manually against the issue sample / reduced repro by replaying the exact failing AOT compiler command and confirming the fixed compiler succeeds where the regressed compiler asserts in inline_method()
• local Mono-style regression validation for src/mono/mono/mini/iltests.il:
  • mono iltests.exe --run-only missing_field_then_inline
  • mono --aot=full,static iltests.exe

This doesn't add a CI runnable test as that would require lot of infra which feels a bit too much for servicing change. (I did try, and hitting this is not simple especially with end-to-end test)

Fixes #129613.

[Copilot]

Pull request overview

This PR narrows Mono AOT typeload-error cleanup in the field-resolution recovery path by consuming additional expected/recoverable metadata errors after the failure is lowered into runtime-throw IR, preventing later inlining assertions from seeing a live cfg->error.

Changes:

• Expand the “recoverable metadata error” cleanup to also clear MONO_ERROR_MISSING_FIELD (in addition to MONO_ERROR_BAD_IMAGE) after HANDLE_TYPELOAD_ERROR in the AOT-only field-resolution failure path.
• Add an explanatory comment describing why cfg->error must be consumed to avoid inline_method()’s mono_error_assert_ok (cfg->error) assertion later in compilation.

[vitek-karas]

@BrzVlad could you please take a look before I bring it up for servicing?

[BrzVlad]

LGTM