adapt native-link-modifier-bundle test no not emit LLVM bitcode

[krasimirgg]

No functional changes intended.
This updates the test case to not embed LLVM bitcode.
This fixes a test failure over at the experimental build of rustc with HEAD LLVM:
https://buildkite.com/llvm-project/rust-llvm-integrate-prototype/builds/11144#01814d0f-a46a-4c19-91cf-41e720edb6f9/684-691.

The issue is that this test uses the system nm, which may not be recent
enough to understand the bitcode produced by rustc when compiled against HEAD LLVM.

Similar to what we did for another test in #94023. Note that we did another thing there, replaced nm with llvm-nm. I decided to not replace nm with llvm-nm here as that was a thing specifically changed in the original PR introducing this test: #95818 (comment).

[rust-highfive]

r? @Mark-Simulacrum

(rust-highfive has picked a reviewer for you, use r? to override)

[Mark-Simulacrum]

specifically changed in the original PR introducing this test

I don't think this change was actually necessary; we should probably use llvm-nm here.

cc @petrochenkov in case you recall why you made that change, but I suspect it wasn't actually a contributor to fixing the test in that PR

[petrochenkov]

we should probably use llvm-nm here

I don't have a strong opinion here.
On one hand this is the only test that used llvm-nm instead of platform-provided nm.
On another hand maybe we should instead migrate all nm tests to llvm-nm if we are building it anyway.

If it passes through CI then it's fine, I guess.

cc @petrochenkov in case you recall why you made that change

I've seen llvm-nm producing an IO error when writing to a pipe where regular nm didn't produce it, but it may be entirely unrelated to CI failures in #95818.

[krasimirgg]

I'll create an alternative of this PR that uses llvm-nm and let it go through CI.