Scalar reduction stores can be uninitialized

[magnatelee]

The semantics of reduction stores admits partial updates, including the "uninitialized" case where the task didn't touch the store at all. However, as reported by this comment, the current implementation requires an accessor be created to make that case work. This shouldn't be a requirement and the postamble should be able to handle this automatically for reduction stores. Note that the same isn't true for write stores, as they must receive updates from the task and reporting the uninitialized case early is actually useful.

[magnatelee]

Closed by #155.