perf(compile): TPC 编译提速——并行 worker 满核 + vast 解析器热点优化

cProfile 分析（206s 全量重编译 TPV）：~163s 为 ProcessPool 等待（16 核机器仅用 8 worker），vast.parse 35.9s（1231 次）为最大 Python 计算热点。

- Phase2Translator：4 处 n_workers 上限 8 → os.cpu_count()（16 核满负荷），并行翻译/llc 编译利用率翻倍
- vast/Tokenizer._PosInfo：isascii 快速路径 + 行索引增量查找（~300 万次调用，免每次 bisect+encode）
- vast/Tokenizer._EmitOp：运算符按首字符分组，58 万次调用从 42 项全表 startswith 降到 2-3 项
- vast/Tokenizer._ScanName：内联 _IsIdentPart（70 万次循环免函数调用）
- Projectrans/Utils：find_reachable 用 deque + 后缀索引替代 pop(0)/全表扫描；topological_sort 用反向邻接表 + heapq 替代每轮 sort+全图扫描

lib/Projectrans/Phase2Translator.py

@@ -417,7 +417,7 @@ class Phase2Translator:
             _vlog().info(f"\n需要翻译 {len(need_translate)} 个文件")
             t_start = time.time()
 
-            n_workers = min(8, len(need_translate))
+            n_workers = min(os.cpu_count() or 4, len(need_translate))
 
             if n_workers > 1 and len(need_translate) > 1:
                 _vlog().info(f"  并行翻译 (workers={n_workers})")
@@ -1643,7 +1643,7 @@ class Phase2Translator:
                 need_compile.append(ll_path)
 
         if need_compile:
-            n_workers = min(os.cpu_count() or 4, len(need_compile), 8)
+            n_workers = min(os.cpu_count() or 4, len(need_compile))
             _vlog().info(f"  并行编译 {len(need_compile)} 个文件 (workers={n_workers})")
             errors = []
             with ThreadPoolExecutor(max_workers=n_workers) as executor:
@@ -2167,7 +2167,7 @@ class Phase2Translator:
                 shared_pickle_path = ''
 
             if shared_pickle_path and os.path.exists(shared_pickle_path):
-                n_workers = min(8, len(need_translate))
+                n_workers = min(os.cpu_count() or 4, len(need_translate))
                 errors = []
                 # 波次并行翻译：按依赖关系分批提交，确保被依赖的文件先翻译完成。
                 # 根因：_TryLoadStructFromStub 扫描 output 目录加载跨模块 struct 定义，
@@ -2392,7 +2392,7 @@ class Phase2Translator:
             return (task, result)
 
         if len(compile_tasks) > 1:
-            n_workers = min(8, len(compile_tasks))
+            n_workers = min(os.cpu_count() or 4, len(compile_tasks))
             with ThreadPoolExecutor(max_workers=n_workers) as executor:
                 futures = {executor.submit(_compile_one_include, task): task for task in compile_tasks}
                 done_count = 0