Mark llvm::Any::TypeId as global in julia.expmap

[vchuravy]

The dynamic linker needs to unify llvm::Any::TypeId across DSOs. In our case
libjulia-codegen and libLLVM.

See https://github.com/llvm/llvm-project/blob/2bc4c3e920ee078ef2879b00c40440e0867f0b9e/llvm/include/llvm/ADT/Any.h#L30

Fixes: #49121

[vchuravy]

@pchintalapudi confirmed offline that this fixes his observed behavior.

Before:

readelf -W --syms usr/lib/libjulia-codegen.so | grep _ZN4llvm3Any6TypeIdIPKNS_6ModuleEE2IdE
  2478: 0000000000178243     1 OBJECT  LOCAL  DEFAULT   15 _ZN4llvm3Any6TypeIdIPKNS_6ModuleEE2IdE

After:

readelf -W --syms usr/lib/libjulia-codegen.so | grep _ZN4llvm3Any6TypeIdIPKNS_6ModuleEE2IdE
  1428: 0000000000178243     1 OBJECT  UNIQUE DEFAULT   15 _ZN4llvm3Any6TypeIdIPKNS_6ModuleEE2IdE
  3928: 0000000000178243     1 OBJECT  UNIQUE DEFAULT   15 _ZN4llvm3Any6TypeIdIPKNS_6ModuleEE2IdE

[vtjnash]

That UNIQUE still seems wrong? Shouldn't it say WEAK to be correct here?

[vchuravy]

It should be correct. We don't pass -fno-gnu-unique when we build libjulia. I think actually the fact that we trigger UNIQUE is the reasons this works (weak should be fine as well, but shrug)

[vtjnash]

Won't it break any other copy of llvm though, since it disables symbol versioning?

[pchintalapudi]

I think for Any's TypeId in particular it should be ok, since all we want is a unique id within the shared library and it's not really state.