make ReinterpretArray more Offset-safe

[adienes]

fixes #44243 (and adds more tests besides)

[adienes]

I tried also adding a wrapper test for an offset array with non-zero offsets. I think the implementation is actually correct (besides what's fixed in this PR) and the tests would pass, but a lot of the tests check equality against a reference array which are no longer valid if the axes differ, so it requires a larger / more annoying refactoring to the test suite here.

[adienes]

one thing to note is that L91 actually makes the defensive first(axes(A, 2)) + ind.j - 1 here unnecessary and we could probably do a bare ind.j. but in the hypothetical case where in the future ReinterpretArray is made more offset-friendly, it would be easy to forget to update that.

another option would be to make the implementation ind.j for now but add a test to catch the hypothetical above

given that this expands the test coverage to two full new wrappers, only changes erroring paths into non-errors (and does not touch any non-erroring paths), and the implementation seems pretty uncontroversial to me, absent protest I'll merge this by the end of the week