Basic support for axis simplification and arbitrary order in iterators

[bluss]

• In some cases we can modify the iterator without changing produced element order - for example changing shape [10, 1] to shape [1, 10]
• In some cases we can merge axes without changing visible element order. Iterating shape [2, 5, 3] as if it is shape [1, 1, 30], if the strides allow
• In some cases we want to use arbitrary order (here it becomes the in-memory order)
• In some cases we don't want to change anything (standard order indexed iterator)

Current state of the PR: groundwork, not actually used in any iterator