MAINT: Move offset calculation code to PyCell

[kngwyu]

Follow up of #1058. Make it cleaner and correct (current code has a bug in that it assumes weakref, dict order).

[sebpuetz]

I had just proposed something similar over at #1060, returning the offset from the start of the struct instead of the end eliminates some additional calculations, too.

#[repr(C)]
pub struct PyCell<T: PyClass> {
    inner: PyCellInner<T>,
    thread_checker: T::ThreadChecker,
    // DO NOT CHANGE THE ORDER OF THESE FIELDS WITHOUT CHANGING PyCell::dict_offset()
    // AND PyCell::weakref_offset()
    dict: T::Dict,
    weakref: T::WeakRef,
}

impl<T: PyClass> PyCell<T> {
    /// Get the offset of the dictionary from the start of the struct in bytes.
    pub(crate) fn dict_offset() -> usize {
        mem::size_of::<Self>() - mem::size_of::<T::Dict>() - mem::size_of::<T::WeakRef>()
    }

    /// Get the offset of the weakref list from the start of the struct in bytes.
    pub(crate) fn weakref_offset() -> usize {
        mem::size_of::<Self>() - mem::size_of::<T::WeakRef>()
    }
}

and

    // __dict__ support
    if !T::Dict::IS_DUMMY {
        type_object.tp_dictoffset = PyCell::<T>::dict_offset() as ffi::Py_ssize_t;
    }

    // weakref support
    if !T::WeakRef::IS_DUMMY {
        type_object.tp_weaklistoffset = PyCell::<T>::weakref_offset() as ffi::Py_ssize_t;
    }

[kngwyu]

Oops, I'm sorry I overlooked that.

returning the offset from the start of the struct instead of the end eliminates some additional calculations, too.

The offset should be zero if the type doesn't have dict or weakref slots, so I think rerunning Option is more convenient in that it doesn't require if offset != type_object.tp_basicsize or so.

[sebpuetz]

As far as I can tell, my proposed solution doesn't need that if condition either? offset is eliminated from initialize_type_object entirely in that. I'll amend the commit over at #1060, should make it easier to discuss then!

[kngwyu]

Closed in favor of #1060