Resolve padding semantics in global dim data spec

Padding support is currently spotty, so there are several places where it needs to be worked on. One area is in the global dim data spec and making it consistent with the protocol's description of padding. For instance, if communication padding is 2, then the global index bounds for a 1D distarray would be something like:

| Process | 0 | 1 |
| --- | --- | --- |
| Padding | (0, 2) | (2, 0) |
| (start, stop) | (0, 13) | (9, 20) |

Note especially that the `stop` of process 0 is greater than the `start` of process 1.

Topic for discussion: how should this be specified in the global dim data? We'd like it to be straightforward, allow full control over how things are specified, and (ideally) minimal, so no data is duplicated.