lusewell

I believe the bug comes from [here](https://github.com/pydata/bottleneck/blob/f2bc792e540908d165143c4930d63d1e7ab806d3/bottleneck/include/iterators.h#L108) assuming that the strides are sorted.

This condition only actualyl needs to be that the last non-zero-strided axis is also the smallest non-zero-strided axis, but this was simpler to write.

Removing the change to iterators.h, the tests still cause a segfault - ie current state on master is broken. The change of reverting the previous release means its quite a...

EG you've reverted a load of bugfixes such as https://github.com/pydata/bottleneck/commit/a8dbfdcc82adbb2778dce4f7b7170de3d27ef23f, and removed the tests that got added for them... Why? Can you readd all of these please?