Unbound or partitioned GlobalRef not allowed in value position on master

[KristofferC]

As an example: https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_date/2025-02/02/Quaternionic.primary.log

Happens in hundreds of packages https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_date/2025-02/02/report.html.

Likely binding partition work, cc @Keno

[Keno]

We'll need to re-evaluate this after #57253, but this is the common failure mode for any sort of bug in the world age tracking, missing edges, etc., because it's the only place that really verifies world age consistency between inference and the runtime system, which we just didn't check before. There's quite likely to be multiple bugs.