Implement expansions and expansion-transformations

[cmdupuis3]

As it stands, there are some features that are included to enable future support for expansion and expansion-transformation functions, but support is incomplete. Notably, although the vanilla loop calls require extents to be passed as an extra argument, the S-dimension extents of the output array are correctly deduced and implemented that way, which makes passing extents look superfluous.

This extent argument should be kept though, because its a good place to put information about T-dimensions.

Also, as originally envisioned, the output array would be sliced along with the input arrays, but this led to persistence issues and was dropped in favor of indexing in the innermost loop only. By coincidence, this should make it easier to decouple how iranks and oranks interact with the rest of the code.

[cmdupuis3]

In the new formalism, the tdim specification has been reworked. Atm it should take structs to specify non-obvious tdims. This work will probably need some rigorous validation to avoid persistence issues, since on the C++ side, a lot of stuff is in danger of going out of scope.