What should we add?

When the cache_pygates feature is enabled, the py_op field in the PackedInstruction struct stores a OnceCell<PyObject> right now. See:

this is used a cached python object for the underlying instruction so we don't repeatedly rebuild the objects when returning to python. However the use of a OnceCell in this field is preventing us from enabling the pygate cache and writing multithreaded transpiler passes as it is not thread-safe. The logical alternative is GilOnceCell which uses the GIL as a lock for a threadsafe version, but it isn't cloneable (which is commented on in the code). The other alternative is OnceLock but that likely will add some overhead to the pygate cache as we're adding an additional locking mechanism.