fix: 修复Python 3.13环境下PyInstaller构建崩溃问题

问题:
- PyInstaller在分析阶段尝试导入torch/transformers等ML库
- 这些库与Python 3.13不兼容，导致"Windows fatal exception: access violation"
- 构建过程无法完成，无法生成exe文件

解决方案:
1. 添加pyi_hooks/pyi_rth_ignore_torch.py运行时hook
   - 在PyInstaller分析阶段阻止torch等模块的导入
   - 这些模块将在运行时动态安装

2. 添加pyi_hooks/hook-exclude-ml.py和hook-paddleocr.py
   - 明确告诉PyInstaller不收集这些ML库

3. 更新build.bat和build.sh
   - 使用--runtime-hook和--additional-hooks-dir
   - 添加--exclude-module参数排除所有ML库

测试:
- 验证所有hook文件语法正确
- 验证项目核心功能可以在不导入torch的情况下正常工作

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

pyi_hooks/pyi_rth_ignore_torch.py

@@ -0,0 +1,47 @@
+"""
+PyInstaller runtime hook to prevent torch import during analysis
+
+This hook runs before PyInstaller's module analysis phase.
+It blocks problematic modules from being imported during the build process.
+"""
+
+import sys
+
+# List of modules to block during PyInstaller analysis
+BLOCKED_MODULES = [
+    'torch',
+    'transformers',
+    'tensorflow',
+    'onnx',
+    'onnxruntime',
+    'sentencepiece',
+    'tokenizers',
+    'diffusers',
+    'accelerate',
+    'datasets',
+    'huggingface_hub',
+    'safetensors',
+]
+
+# Override the import mechanism to block these modules
+class BlockModuleImport:
+    """Meta path importer to block specific modules during PyInstaller build"""
+
+    def find_module(self, fullname, path=None):
+        if fullname in BLOCKED_MODULES or any(
+            fullname.startswith(m + '.') for m in BLOCKED_MODULES
+        ):
+            # Return self to handle the import (will raise ImportError)
+            return self
+        return None
+
+    def load_module(self, fullname):
+        raise ImportError(
+            f"Module '{fullname}' is excluded from PyInstaller build. "
+            f"It will be installed at runtime if needed."
+        )
+
+# Install the blocker at the start of PyInstaller analysis
+if '_MEIPASS' not in sys.__dict__:
+    # Only during build time (not when running the frozen app)
+    sys.meta_path.insert(0, BlockModuleImport())