[ICE]: Missing incremental LTO work product panics instead of rebuilding

Reproduction

The minimal self-verifying reproducer is available at:

https://github.com/ashi009/rustc-missing-incremental-work-product-repro

git clone https://github.com/ashi009/rustc-missing-incremental-work-product-repro.git
cd rustc-missing-incremental-work-product-repro
rustup run nightly bash repro.sh

The script populates an incremental cache, removes one cached .pre-lto.bc
work product, and repeats the identical compilation. It exits successfully only
when it observes the expected failed to open LTO bitcode file ICE.

Since incremental state is a cache, I expected rustc to discard the unusable
session and rebuild successfully.

This provides a deterministic reproducer for #94943, which was closed because
no reproducer was available.

Meta

Reproduces with stable:

rustc 1.98.0 (88d9e12ae 2026-08-18)
host: aarch64-apple-darwin

Also reproduces with current nightly:

rustc 1.100.0-nightly (a36d05efa 2026-09-09)

Error output

thread 'coordinator' panicked at compiler/rustc_codegen_ssa/src/back/lto.rs:56:13:
failed to open LTO bitcode file `.../s-...-working/....pre-lto.bc`:
No such file or directory (os error 2)

error: the compiler unexpectedly panicked. This is a bug

rustc_codegen_ssa::back::lto::SerializedModule::from_file
rustc_codegen_llvm::back::lto::thin_lto
rustc_codegen_llvm::LlvmCodegenBackend::run_thin_lto
rustc_codegen_ssa::back::write::do_thin_lto