gh-159312: Avoid CPU attribution to suspended thread states

[pablogsal]

CPU mode can attribute samples to a suspended stack when another thread state uses the same native thread. For an unbound state, check whether a bound state on that native thread has Python frames before marking it on CPU. Preserve attribution when no bound stack is found, including native work after a callback deletes or empties its temporary state. Bound the lookup and keep the sample if remote reads fail.

The normal path adds one flag check, with no new allocation or remote read. The regression uses the existing test helper. External-inspection and profiling tests pass on regular and free-threaded debug builds, with 547 tests on each. A small Linux debug benchmark of ordinary sampling showed median changes of -1.9% for one thread and +0.6% for eight, within the observed run-to-run variation.

• Issue: profiling.sampling CPU mode may attribute on-CPU samples to detached tstates when sharing native_thread_id #159312