bpo-33895: Relase GIL while calling functions that acquire Windows loader lock (GH-7789)

LoadLibrary, GetProcAddress, FreeLibrary and GetModuleHandle acquire the system loader lock. Calling these while holding the GIL will cause a deadlock on the rare occasion that another thread is detaching and needs to destroy its thread state at the same time.

Python/dynload_win.c

@@ -218,8 +218,10 @@ dl_funcptr _PyImport_FindSharedFuncptrWindows(const char *prefix,
         /* We use LoadLibraryEx so Windows looks for dependent DLLs
             in directory of pathname first. */
         /* XXX This call doesn't exist in Windows CE */
+        Py_BEGIN_ALLOW_THREADS
         hDLL = LoadLibraryExW(wpathname, NULL,
                               LOAD_WITH_ALTERED_SEARCH_PATH);
+        Py_END_ALLOW_THREADS
 #if HAVE_SXS
         _Py_DeactivateActCtx(cookie);
 #endif
@@ -298,11 +300,15 @@ dl_funcptr _PyImport_FindSharedFuncptrWindows(const char *prefix,
                              "Module use of %.150s conflicts "
                              "with this version of Python.",
                              import_python);
+                Py_BEGIN_ALLOW_THREADS
                 FreeLibrary(hDLL);
+                Py_END_ALLOW_THREADS
                 return NULL;
             }
         }
+        Py_BEGIN_ALLOW_THREADS
         p = GetProcAddress(hDLL, funcname);
+        Py_END_ALLOW_THREADS
     }
 
     return p;

Python/sysmodule.c

@@ -1159,7 +1159,9 @@ sys_getwindowsversion_impl(PyObject *module)
     // We need to read the version info from a system file resource
     // to accurately identify the OS version. If we fail for any reason,
     // just return whatever GetVersion said.
+    Py_BEGIN_ALLOW_THREADS
     hKernel32 = GetModuleHandleW(L"kernel32.dll");
+    Py_END_ALLOW_THREADS
     if (hKernel32 && GetModuleFileNameW(hKernel32, kernel32_path, MAX_PATH) &&
         (verblock_size = GetFileVersionInfoSizeW(kernel32_path, NULL)) &&
         (verblock = PyMem_RawMalloc(verblock_size))) {