Add map_mut and map_axis_mut

[LukeMathWalker]

Following #452 I tried to implement map_mut and map_axis_mut.

There is unsafe code in multiple spots and my familiarity with the overall architecture of ndarray (and Rust in general) is still lacking, so I'd like to understand if some of my choices might be calling for disaster in disguise:

• I have implemented TrustedIterator for IterMut and slide::IterMut - was there any reason this was not already in place (apart from not having been needed so far)?
• to avoid issues with the borrow checker in the else branch I have modified

if let Some(slc) = as_slice_memory_order_mut() {
    [...]
} else {
    [...]
}

into

if self.is_contiguous() {
    let slc = self.as_slice_memory_order_mut().unwrap();
    [...]
} else {
    [...]
}

which seemed reasonable looking at as_slice_memory_order_mut implementation:

    /// Return the array’s data as a slice if it is contiguous,
    /// return `None` otherwise.
    pub fn as_slice_memory_order_mut(&mut self) -> Option<&mut [A]>
        where S: DataMut
    {
        if self.is_contiguous() {
            self.ensure_unique();
            unsafe {
                Some(slice::from_raw_parts_mut(self.ptr, self.len()))
            }
        } else {
            None
        }
    }

@jturner314 @rcarson3

[bluss]

Nice, just small things to fix

[bluss]

Please avoid formatting changes outside the code you are editing.

[LukeMathWalker]

My vim settings removes trailing whitespaces on save... Got to change it when I am working on ndarray :P

[jturner314]

I have Emacs set up the same way. Fwiw, I've found it useful to have the "save" keybinding set to remove trailing whitespace, but preserve the ability to call the save-buffer command directly if desired to save without removing whitespace.

Anyway, I've submitted PR #464 to remove all the trailing whitespace in ndarray so that we won't have to worry about this issue anymore.

For the purpose of this PR, though, @bluss is right that the PR should not reformat lines unrelated to the changes.

[LukeMathWalker]

Yes, I was thinking about doing something similar to that in my .init.vim.
Until #464 lands I guess I'll be restoring whitespaces once the code has been approved and it is just waiting to be merged - final clean-up, so to speak.

[bluss]

This redundant is_contiguous call is unfortunate. The strides.clone call can move inside the first if branch though, so that's nice.

[bluss]

This A: Clone here is unwanted

[bluss]

The 'a here is good.

[bluss]

The 'a parameter now has no use and can be removed. (It having no use unlike the &self version of map, is correct.)

[bluss]

Hm.. wait a minute, that's not right, using 'a the same way as the &self map would be good.

[bluss]

This looks good!

[LukeMathWalker]

Thanks for reviewing @bluss!
I should have implemented/fixed all lines you pointed to (as well as restoring trailing whitespaces). :D

[bluss]

Thanks a lot for this!