Fix: avoid -C prefer-dynamic for test harnesses of proc-macro crates

[tommantonclery]

Fixes #16000.

When running cargo test --no-run on a proc-macro crate with a dev-dependency on a dylib, Cargo was incorrectly passing -C prefer-dynamic to the test harness.

That caused rustc to try to link both rlib and dylib forms of std and its transitive crates, resulting in errors like:

error: cannot satisfy dependencies so `std` only shows up once

What changed

• Adjusted the logic in core/compiler so that -C prefer-dynamic is only applied when compiling an actual proc-macro crate.
• Test harnesses (even for proc-macro crates) no longer receive -C prefer-dynamic.

Why

This resolves the duplicate std/core/alloc linkage errors and makes cargo test --no-run consistent with cargo build.

[rustbot]

r? @epage

rustbot has assigned @epage.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[Firestar99]

Just noticed this PR fixing my issue, could we get it reviewed and merged?

[weihanglo]

Haven't yet thought hard on this, but it at leaste requires a test to make sure it fixes the real issue

BTW, we usually follow a variant of atomic commit pattern:

1. Commit a test that asserts the current buggy behavior (passes).
2. In the next commit, fix the bug and update the test/snapshot.

Every commit passes, and the test/snapshot diff shows the behavior change.

View changes since this review