cc-test: add loongarch64 assembly shim

[tangaac]

Problem

dev-tools/cc-test's build.rs compiles src/<arch>.S based on the target architecture. aarch64.S, armv7.S, i686.S, riscv64gc.S, and x86_64.S are all present, but there is no loongarch64.S. On loongarch64-unknown-linux-gnu (stable Rust target since 1.75) this fails with:

cc1: fatal error: src/loongarch64.S: No such file or directory

The cc crate itself already declares support for LoongArch64 (loongarch64-unknown-linux-gnu => loongarch64-linux-gnu in src/lib.rs), so this is purely a missing dev-tool shim.

Fix

Add dev-tools/cc-test/src/loongarch64.S, modeled on riscv64gc.S, exporting both asm and _asm symbols. Verified on real LoongArch64 hardware: cargo build succeeds and cargo test passes 9/9 — the asm_here test asserts the shim returns 7.

[NobodyXu]

Thank you!

Would you like to add a CI job to compile the test on loongarch?

It's ok if you don't want to and I can get this merged

[tangaac]

I dropped the check-loongarch64 CI job — cross-linking the test suite on an x86_64 runner hits an unrelated problem. build.rs has a cfg(target_os = "linux") block (for the static-libstdc++ link test) that runs the host g++ --print-file-name=libstdc++.a and symlinks the host libstdc++.a into OUT_DIR. On an x86_64 runner that is an x86_64 archive, and rustc embeds its objects (e.g. del_ops.o) into the test rlib, so the loongarch64 linker fails with Relocations in generic ELF (EM: 62). This only surfaces on a linux cross target — the existing cross jobs are Apple/Windows, so this path was never exercised.

So this PR is just the shim for now. It is verified on a real LoongArch64 machine: the assembly compiles and asm_here passes.

[NobodyXu]

Thanks, merged