feat(funcattrs): 支持 stackrealign 函数属性

入口函数（_start）由 bootloader jmp 进入（rsp≡0 mod 16），llc 按 call 约定
假设入口 rsp≡8 → 帧整体偏 8 → 所有调用 rsp≡8 → 变参函数保存 xmm 的
movaps #GP。新增 t.attr.stackrealign → LLVM stackrealign 属性，序言动态
重对齐栈，与入口约定无关。

App/lib/core/Gen/FuncGenBase.vp

@@ -142,6 +142,12 @@ class FuncGenBase(CallGenBase.CallGenBase):
             llvm_name = "nounwind"
         elif string.strcmp(attr_name, "noredzone") == 0:
             llvm_name = "noredzone"
+        elif string.strcmp(attr_name, "stackrealign") == 0:
+            # 动态重对齐栈（LLVM 序言 and rsp,-16）。入口函数（_start）由 bootloader
+            # jmp 进入（rsp≡0 mod 16），而 llc 按 call 约定假设入口 rsp≡8 → 帧整体
+            # 偏 8 → 所有调用 rsp≡8 → 变参函数保存 xmm 的 movaps #GP。stackrealign
+            # 使序言动态对齐，与入口约定无关（sched started 后 movaps #GP 根因）。
+            llvm_name = "stackrealign"
         elif string.strcmp(attr_name, "willreturn") == 0:
             llvm_name = "willreturn"
         elif string.strcmp(attr_name, "mustprogress") == 0:
@@ -283,6 +289,57 @@ class FuncGenBase(CallGenBase.CallGenBase):
                 return 1
         return 0
     # ============================================================
+    # _resolve_creturn_ret_ty - @c.CReturn(ty1, ty2, ...) → 多返回 struct 类型
+    #
+    # 对齐 TPC DeclarationGenerator：CReturn 声明函数返回多个值（如
+    # GetMemoryStats @c.CReturn(t.CUInt64T, t.CUInt64T, t.CUInt64T) →
+    # {i64, i64, i64}）。pre_scan 前向声明与 translate_function_def 定义
+    # 两阶段必须一致，否则 declare i32 + define {i64,i64,i64} → llc
+    # "ret doesn't match function result type"。
+    # ============================================================
+    def _resolve_creturn_ret_ty(self, pool: memhub.MemBuddy | t.CPtr,
+                                fd: ast.FunctionDef | t.CPtr,
+                                imported_modules: str,
+                                from_imports: str,
+                                trans: HT.Translator | t.CPtr) -> llvmlite.LLVMType | t.CPtr:
+        """解析 @c.CReturn 装饰器为 struct 返回类型；无 CReturn 返回 None"""
+        if fd is None or fd.decorator_list is None:
+            return None
+        cr_head: llvmlite.ParamNode | t.CPtr = None
+        cr_last: llvmlite.ParamNode | t.CPtr = None
+        cr_count: int = 0
+        cr_dn: t.CSizeT = len(fd.decorator_list)
+        for cdi in range(cr_dn):
+            deco_cr: ast.AST | t.CPtr = fd.decorator_list.get(cdi)
+            if deco_cr is None or deco_cr.kind() != ast.ASTKind.Call:
+                continue
+            deco_call_cr: ast.Call | t.CPtr = (ast.Call | t.CPtr)(deco_cr)
+            if deco_call_cr.func is None or deco_call_cr.func.kind() != ast.ASTKind.Attribute:
+                continue
+            deco_fa_cr: ast.Attribute | t.CPtr = (ast.Attribute | t.CPtr)(deco_call_cr.func)
+            if deco_fa_cr.attr is None or string.strcmp(deco_fa_cr.attr, "CReturn") != 0:
+                continue
+            cr_args: list[ast.AST | t.CPtr] | t.CPtr = deco_call_cr.args
+            if cr_args is None:
+                continue
+            cr_an: t.CSizeT = len(cr_args)
+            for cai in range(cr_an):
+                cr_aty: llvmlite.LLVMType | t.CPtr = self.Gen.TypeGenResolve.resolve_annotation_type(
+                    pool, cr_args.get(cai), imported_modules, from_imports, trans, 0)
+                if cr_aty is None:
+                    continue
+                cr_pn: llvmlite.ParamNode | t.CPtr = llvmlite.new_param_node(pool, cr_aty)
+                if cr_pn is not None:
+                    if cr_head is None:
+                        cr_head = cr_pn
+                    else:
+                        cr_last.Next = cr_pn
+                    cr_last = cr_pn
+                    cr_count += 1
+        if cr_head is not None and cr_count > 0:
+            return llvmlite.Struct(pool, cr_head, cr_count, None)
+        return None
+    # ============================================================
     # 创建 LLVM 函数（简单版本，仅声明）
     # ============================================================
     def create_function(self, pool: memhub.MemBuddy | t.CPtr,
@@ -316,7 +373,7 @@ class FuncGenBase(CallGenBase.CallGenBase):
                         param_ty: llvmlite.LLVMType | t.CPtr = i32_ty
                         if arg.annotation is not None:
                             resolved: llvmlite.LLVMType | t.CPtr = self.Gen.TypeGenResolve.resolve_annotation_type(
-                                pool, arg.annotation, None, None, None)
+                                pool, arg.annotation, None, None, None, 0)
                             if resolved is not None:
                                 param_ty = resolved
                         pname: str = pool.alloc(32)