API: Make datetime64 timezone naive

Fixes GH3290

With apologies to mwiebe, this rips out most of the time zone parsing from
the datetime64 type.

I think we mostly sorted out the API design in discussions last year, but
I'll be posting this to the mailing list shortly to get feedback.

Old behavior:

	# string parsing and printing defaults to your local timezone :(
    >>> np.datetime64('2000-01-01T00')
    numpy.datetime64('2000-01-01T00:00-0800','h')

New behavior:

    # datetime64 is parsed and printed as timezone naive
    >>> np.datetime64('2000-01-01T00')
    numpy.datetime64('2000-01-01T00','h')

    # you can still supply a timezone, but you get a deprecation warning
    >>> np.datetime64('2000-01-01T00Z')
    DeprecationWarning: parsing timezone aware datetimes is deprecated; this
    will raise an error in the future
    numpy.datetime64('2000-01-01T00','h')

Fixes gh-7010

Fixes GH2039

This function provides a much more intuitive interface than `np.rollaxis`,
which has a confusing behavior with the position of the `start` argument:
http://stackoverflow.com/questions/29891583/reason-why-numpy-rollaxis-is-so-confusing

It was independently suggested several times over the years after discussions
on the mailing list and GitHub (GH2039), but never made it into a pull request:
https://mail.scipy.org/pipermail/numpy-discussion/2010-September/052882.html

My version adds support for a sequence of axis arguments. I find this behavior
to be very useful. It is often more intuitive than supplying a list of
arguments to `transpose` and also nicely generalizes NumPy's existing axis
manipulation routines, e.g.,

    def transpose(a, order=None):
        if order is None:
            order = reversed(range(a.ndim))
        return moveaxes(a, order, range(a.ndim))

    def swapaxes(a, axis1, axis2):
        return moveaxes(a, [axis1, axis2], [axis2, axis1])

    def rollaxis(a, axis, start=0):
        if axis < start:
            start -= 1
        return moveaxes(a, axis, start)

This looks more conventionnal

* Change ".. function::" -> ".. method::"
* Remove "self" argument
* Change "self" to "obj" in __array_finalize__

The user guide was missing a quick tutorial --- the basics.* stuff is
somewhat too complex already.

The "building numpy" instructions also should not be "introductory
material".