ci(windows): drop rustc's stale libwinpthread so C++ proc-macros load

[nebasuke]

Workaround for #589 (the load-time sequel to #550).

Problem

Slang Tests → Windows is deterministically red on main since #557 merged (runs 29910202004, 29912409528):

error[E0463]: can't find crate for `melior_macro`
error: could not compile `melior` (lib) due to 1 previous error

The melior rev bump in #557 is the trigger, not the cause — it invalidated the cargo cache entry and forced the first fresh melior compile (= first proc-macro load) since the #550 shared-libstdc++ workaround landed. melior-macro's source is byte-identical across the two revs, and no good melior artifact can ever be cached while the load fails, so the leg stays red without this fix.

Root cause (confirmed by PE import/export analysis — full trail in #589)

melior-macro is a proc-macro that links LLVM C++ (via tblgen). Since the #550 workaround switched test builds to shared libstdc++, the built melior_macro.dll needs GCC 16's libstdc++-6.dll at load time.

rustc loads proc-macros via libloading → LoadLibraryExW(path, 0) → the default Windows DLL search order, in which the application directory — rustc's sysroot bin/ — beats PATH. The windows-gnu dist ships a stale libwinpthread-1.dll there, and GCC 16's libstdc++-6.dll imports three mingw-w64 v14 64-bit-time symbols the stale copy doesn't export:

GCC 16 libstdc++-6.dll imports from winpthread	rust 1.97.1's bundled libwinpthread-1.dll
clock_gettime64	❌ absent
nanosleep64	❌ absent
pthread_cond_timedwait64	❌ absent

The dependency chain fails to load and rustc surfaces it as the generic E0463: can't find crate. PATH-based fixes cannot work — the app dir always wins.

Fix

Delete the stale DLL from the sysroot bin/ in build-toolchain (next to the existing #550 workaround, same guards), so the dependency falls through to MSYS2's mingw-w64 v14 copy via PATH:

rm -fv "$(rustc --print sysroot)/bin/libwinpthread-1.dll"

Verified safe: nothing in the sysroot bin/ imports winpthread — rustc.exe, rustdoc.exe, rustc_driver-*.dll, and std-*.dll all checked (they depend only on KERNEL32/msvcrt/ntdll/libgcc). rust-lld.exe does need it, but has its own copies under rustlib/<triple>/bin/ (rust-lang/rust#128876), untouched here. The libgcc half of the chain needs no intervention: all 14 symbols GCC 16's libstdc++ imports from libgcc_s_seh-1.dll are exported by rust's bundled copy. -f makes the step self-nooping (and deletable) once rust ships mingw-w64 ≥ v14.

Prior art — same class, same fix shape

• Incompatibility between rustc's libgcc_s_dw2-1.dll and MSYS2 mingw-w64-i686-gcc 11.3 rust-lang/rust#99534 (via rustup's libgcc_s_dw2-1.dll breaks gcc under i686-pc-windows-gnu rust-lang/rustup#3045): rustc's stale bundled libgcc_s_dw2-1.dll broke MSYS2 gcc ≥ 11.3; reporter's workaround was deleting the DLL; fixed upstream by a bundled-mingw bump (Upgrade mingw-w64 on CI rust-lang/rust#100178).
• Executables fail to launch when build with latest MSYS2/MinGW due to outdated libwinpthread-1.dll ddnet/ddnet#10203 and a ruby-core report: identical stale-winpthread failures outside Rust, both resolved by removing/replacing the stale DLL.

An upstream report against rust-lang/rust is being filed in parallel (the bin/ copy is imported by nothing there and shadows newer runtimes during in-process loads); this workaround stays until a fixed toolchain reaches our pinned rust-toolchain.toml. The structural fix that makes it deletable on our side — getting tblgen out of macro-expansion time in the melior fork — is tracked in #589.

Verification

Verified in run 29923864614: the Windows leg passes — the cached GCC-16-linked melior_macro.dll (the exact artifact that failed on main) loads, melior compiles, and the slang suite runs green, with this one-file change as the only difference from the failing configuration. The run's overall red comes from solx-mlir :: conditional.sol failing on Linux ARM64 + macOS x86 — a pre-existing main breakage from #563, unrelated to this Windows-gated change; tracked in #591.

[github-actions]

Coverage Summary

Crate	Line Coverage	Function Coverage
solx	🟢 94.7%	🟡 68.0%
solx-benchmark-converter	🔴 0.0%	🔴 0.0%
solx-codegen-evm	🟢 88.8%	🟢 83.2%
solx-compiler-downloader	🔴 0.0%	🔴 0.0%
solx-core	🟢 96.6%	🟢 91.0%
solx-dev	🔴 2.4%	🔴 3.0%
solx-evm-assembly	🟡 70.6%	🟡 66.2%
solx-solc-test-adapter	🔴 1.7%	🔴 2.1%
solx-standard-json	🟢 95.4%	🟢 91.5%
solx-tester	🔴 38.2%	🔴 35.9%
solx-utils	🟢 84.4%	🟢 85.1%
solx-yul	🟡 78.9%	🟡 71.3%
Total	🟡 53.9%	🔴 47.1%

Codecov Report | HTML Report | Workflow Run

[Copilot]

Pull request overview

Adjusts the Windows CI toolchain bootstrap to avoid a deterministic proc-macro load failure in Slang tests caused by rustc’s sysroot bin/ shadowing MSYS2’s newer MinGW runtime DLLs.

Changes:

• Adds a Windows-only step in the shared build-toolchain composite action to remove rustc’s stale libwinpthread-1.dll from the sysroot bin/, allowing DLL resolution to fall through to MSYS2 via PATH.
• Documents the rationale and why the step is expected to become a no-op once Rust ships a mingw-w64 v14+ bundled runtime.