gh-132070: Use _PyObject_IsUniquelyReferenced in unicodeobject

[corona10]

• Issue: data race when resizing unicode objects #132070

[corona10]

@kumaraditya303 @colesbury
I continued working on Kumar's PR #132091. In the new change, more states of unicode_modifiable are observed in free-threading builds. So I added a fallback to use resize_copy when unicode_modifiable fails, but I'm not sure if this is the right approach.
But the test failure issues are resolved.

[serhiy-storchaka]

LGTM.

[colesbury]

I looked into the case that requires resize_copy: We have an io.StringIO, which is used first by one thread and then by a second thread. (It's not used by two threads simultaneously). The io.StringIO has the only reference to a PyUnicodeWriter, which in turn has the only reference to unicode object that is treated as a buffer.

The _PyObject_IsUniquelyReferenced() check is too conservative in this case because the unicode object is owned by a different thread. The unicode object is actually safe to modify, we just can't tell that from _PyObject_IsUniquelyReferenced().