Fix weights_only for BUILD instructions for user allowlisted objects with __slots__ (#138936)

Previously `BUILD` instruction missed handling for `__slots__`. **This only applies for things allowlisted via `add_safe_globals`/`safe_globals` that use slots.**

### Background
When does pickle serialize a `BUILD` instruction? When `state` is not `None` and `state_setter` is `None` [[link](c5b99f5c2c/Lib/pickle.py (L765))]. In this case, the docs tell us that either `__setstate__` or a `__dict__` update will be performed [[link](https://github.com/python/cpython/blob/3.13/Lib/pickletools.py#L1984)]

`__reduce__`/`__reduce_ex__` are expected to return tuples of length 2 to 6 where `state` is the 3rd argument. When user doesn't patch `__reduce__` but patches `__setstate__`/`__getstate__`, state will be what is yielded by `__getstate__`

Note the return type for [`__getstate__` ](https://docs.python.org/3/library/pickle.html#object.__getstate__)

- For a class that has no instance [`__dict__`](https://docs.python.org/3/reference/datamodel.html#object.__dict__) and no [`__slots__`](https://docs.python.org/3/reference/datamodel.html#object.__slots__), the default state is None.
- For a class that has an instance [`__dict__`](https://docs.python.org/3/reference/datamodel.html#object.__dict__) and no [`__slots__`](https://docs.python.org/3/reference/datamodel.html#object.__slots__), the default state is `self.__dict__`.
- For a class that has an instance [`__dict__`](https://docs.python.org/3/reference/datamodel.html#object.__dict__) and [`__slots__`](https://docs.python.org/3/reference/datamodel.html#object.__slots__), the default state is a tuple consisting of two dictionaries: `self.__dict__`, and a dictionary mapping slot names to slot values. Only slots that have a value are included in the latter.
- For a class that has [`__slots__`](https://docs.python.org/3/reference/datamodel.html#object.__slots__) and no instance [`__dict__`](https://docs.python.org/3/reference/datamodel.html#object.__dict__), the default state is a tuple whose first item is None and whose second item is a dictionary mapping slot names to slot values described in the previous bullet.

see handling in pickle code c5b99f5c2c/Lib/pickle.py (L1846-L1867)

Before this PR, we didn't account for the fact that when `__setstate__` is not defined, `state` might be a tuple so this would fail

```python
from dataclasses import dataclass

# Define the dataclass
@dataclass
class MyDataClass:
    __slots__ = ["x", "y"]
    x: int
    y: str
# Create an instance of the dataclass
my_data = MyDataClass(x=2, y=3)
# Save the dataclass to a file
torch.save(my_data, "my_data.pt")
with torch.serialization.safe_globals([MyDataClass]):
    loaded_my_data = torch.load("my_data.pt", weights_only=True)
# AttributeError: 'MyDataClass' object has no attribute '__dict__'
```

Pull Request resolved: https://github.com/pytorch/pytorch/pull/138936
Approved by: https://github.com/malfet

torch/_weights_only_unpickler.py

@@ -292,6 +292,13 @@ class Unpickler:
                 elif type(inst) in _get_user_allowed_globals().values():
                     if hasattr(inst, "__setstate__"):
                         inst.__setstate__(state)
+                    elif hasattr(inst, "__slots__"):
+                        # if slots are defined, state will be a tuple (state, slotstate)
+                        state, slotstate = state
+                        for k, v in slotstate.items():
+                            setattr(inst, k, v)
+                        if state:
+                            inst.__dict__.update(state)
                     else:
                         inst.__dict__.update(state)
                 else: