This PR is split off of #129108. Ping @kumaraditya303

EDIT: Removed, unnecessary critical sections in bytearrayiter_next and bytearrayiter_reduce. Really no point having them in this PR, will add them back in bytearray PR after this one is merged.

@kumaraditya303 I thought you would do the iterator first and backport it so removed an unnecessary critical section here if bytearray not safe itself. If you don't intend to backport this then I will just add it back here, but otherwise it doesn't make sense (though harmless) unless you backport bytearray as well (which I am not suggesting). For reference, the critical section is:

static PyObject *
bytearrayiter_next(PyObject *self)
{
    bytesiterobject *it = _bytesiterobject_CAST(self);
    int val;

    assert(it != NULL);
    Py_ssize_t index = FT_ATOMIC_LOAD_SSIZE_RELAXED(it->it_index);
    if (index < 0) {
        return NULL;
    }
    PyByteArrayObject *seq = it->it_seq;
    assert(PyByteArray_Check(seq));

    Py_BEGIN_CRITICAL_SECTION(seq);
    if (index < PyByteArray_GET_SIZE(seq)) {
        val = (unsigned char)PyByteArray_AS_STRING(seq)[index];
    }
    else {
        val = -1;
    }
    Py_END_CRITICAL_SECTION();

    if (val == -1) {
        FT_ATOMIC_STORE_SSIZE_RELAXED(it->it_index, -1);
#ifndef Py_GIL_DISABLED
        it->it_seq = NULL;
        Py_DECREF(seq);
#endif
        return NULL;
    }
    FT_ATOMIC_STORE_SSIZE_RELAXED(it->it_index, index + 1);
    return _PyLong_FromUnsignedChar((unsigned char)val);
}

Though can also just use bytearray_getitem here, but I don't like calling an unnecessary function where can just read a byte instead.

I don't intend to backport the thread safety of bytearray as change is large and there isn't much point in backporting the iterator while bytearray isn't thread safe.

@colesbury Do you think we should backport the thread safety of bytearray?

Yes, it is not safe to read data without acquiring critical section of the seq, you need to add back the critical section, the change to use atomic makes sense.

I probably wouldn't backport the changes.