gh-149816: fix concurrent kwargs growth in _thread.start_new_thread

[YvesDup]

IMO, to fix this issue, there are 2 options:

1. in the PyStack_UnpackDict function located in /Objects/call.c, wraps the PyDict_Next loop in a critical section to prevent concurrent mutations of the same dict.

2. in the /Modules/_threadmodule.c, duplicate the kwargs dict in the thread_run function, before calling the thread's main function. So that the thread will be working with a private copy of the kwargs dict.

In the option 1, the PyStack_UnpackDict function is called many times, from PyObject_VectorcallDict via _PyObject_VectorcallDictTstate, PyVectorcall_Call via _PyVectorcall_Call or _PyEvalFramePushAndInit_Ex. This change is not necessary for these functions and, furthermore, would have an impact on their performance.

I am in favor of option 2, as it has a more limited scope and shouldn't have a performance impact
on regular builds, while still fixing the issue for FT builds.

• Issue: 22 free-threading race conditions #149816