Place CPython thread types in their defining header’s Go file

[Copilot]

TestPython passed despite placing X_PyTssT in pythread.go instead of cpython-pythread.go.

• File routing: Route forward class and struct declarations to the header containing their definition.
• Python goldens: Move X_PyTssT into cpython-pythread.go; keep its PyTssT alias in pythread.go.

// cpython-pythread.go
type X_PyTssT struct { ... }

// pythread.go
type PyTssT = X_PyTssT

• Fixes fix(tool): TestPython bugfix #924

[fennoai]

Review summary

The core fix in cl/compile.go is small, well-targeted, and follows the existing idioms in the cl/ package (IsCursorDefinition() == 0, Definition(), IsNull() == 0, clang.PresumedFile). For a forward-declared class/struct it resolves the file/package from the definition's location instead of the forward declaration's location, which correctly fixes the misattribution the generated fixture change (X_PyTssT moving from pythread.go to cpython-pythread.go) demonstrates. The null-definition guard is correct and the change is behaviorally conservative. Build passes.

No security concerns. The go.mod/go.sum pruning in tool/_testcpp/llvm-22.1.8-support/ leaves a consistent module graph (removed llarhub/libcxx and llarhub/llvm-c are no longer referenced by any Go import there). Generated fixtures are mechanical and out of scope.

Findings below are non-blocking; the main one worth a decision is whether enum/union forward declarations should get the same treatment.

[fennoai]

[P2] Same forward-decl misattribution likely applies to enum/union

This switch only special-cases Cursor_ClassDecl/Cursor_StructDecl. Enum and union declarations short-circuit on the exact same condition elsewhere (enum.go:65 and union.go:67 both do if decl.IsCursorDefinition() == 0 { return }), so an enum/union forward-declared in file A and defined in file B can be attributed to the wrong file by the same mechanism this PR fixes for class/struct. Enum forward declarations with a fixed underlying type (enum E : int;) are legal and common C++11.

If the broader fix is intended, consider also handling lc.Cursor_EnumDecl (and lc.Cursor_UnionDecl). If class/struct is the deliberate scope, a short comment explaining why would prevent a future reader from assuming it was an oversight. (Typedefs are correctly excluded — they are not a forward/definition pair.)

[xushiwei]

@copilot Fix above review comments. And revert any changes in tool/_testcpp/llvm-22.1.8-support (it makes some other test cases failed).

[Copilot]

@copilot Fix above review comments. And revert any changes in tool/_testcpp/llvm-22.1.8-support (it makes some other test cases failed).

Restored the LLVM fixture dependency metadata; the affected TestLLVM_* cases now pass. The fix is in eea6704. I also removed toolchain-installer changes to the root module files in 9b67a2e.