CGU reuse tracking used by incr. comp. testing is inaccurate wrt ThinLTO CGU reuse

[michaelwoerister]

Since #115964, CGU reuse tracking does not report "post-lto" reuse when the compiler discovers that the cached post-LTO version of a CGU can be re-used during the ThinLTO phase in the LLVM backend.

Due to this some tests in https://github.com/rust-lang/rust/tree/cda4736f1eaad8af6f49388baa9b7e480df8e329/tests/incremental/thinlto state "pre-lto" even though we actually expect the "post-lto" artifacts to be reusable. We should look into

• restoring accurate reporting for the LLVM backend, and
• trying to make the tests in question compatible with other backends that don't have ThinLTO.