impr: add T: Send for AtomicLazyCell Sync impl

Add a `T: Send` to the existing `T: Sync` restriction on the `AtomicLazyCell`
`Sync` impl.
This is necessary because filling an `AtomicLazyCell` only requires a reference
to the cell (i.e. `&AtomicLazyCell`), which means that an empty
`&AtomicLazyCell<T>` can be passed to a child thread (provided that
`AtomicLazyCell<T>: Sync` is satisfied), and filled on the child thread.
The filled value would then be on the stack of, and destroyed by, the
parent thread.

Kudos to dbaupp on Reddit for this recommendation.

Closes #67.
BREAKING CHANGE: the `AtomicLazyCell` `Sync` impl now has a `<T: Sync + Send>`
    constraint instead of a `<T: Sync>` constraint.

    To migrate your code, remove any `AtomicLazyCell<T>` instances where
    `<T: Sync + !Send>`, as this usecase is no longer supported.

src/lib.rs

@@ -223,7 +223,7 @@ impl<T: Copy> AtomicLazyCell<T> {
     }
 }
 
-unsafe impl<T: Sync> Sync for AtomicLazyCell<T> {}
+unsafe impl<T: Sync + Send> Sync for AtomicLazyCell<T> {}
 
 unsafe impl<T: Send> Send for AtomicLazyCell<T> {}