Change default value of always_2d to False

[mgeier]

While I'm at suggesting breaking changes (see #132), I thought I might as well suggest another one ...

We discussed this already to quite some length in #16, but while I'm using it, I get the feeling that we may have chosen the wrong pill.

Arguments for always_2d=True:

• it's consistent in itself (everything is 2D, even mono signals)
• it's what you normally want in generic code (where you don't know if you get a mono or multi-channel file)

Arguments for always_2d=False:

• it's consistent with most of NumPy/SciPy (one-channel signals are 1D arrays)
• it's what you normally want in an interactive session (where you normally know if you are working on mono or multi-channel files)

I think we should simplify the interactive case, where being concise and the amount of typing matters most.
In the non-interactive case, it would even make the code more explicit and therefore easier to read, if people are forced to explicitly add always_2d=True.
This would be the equivalent to (but of course much simpler than) manually checking for the dimensions of an input array and reshaping it to the expected dimensionality, which is for sure done in many code bases.

There are several functions (like matplotlib.pyplot.specgram()) which only work on one-dimensional signals.
For those it would clearly be more practical to use always_2d=False.

There are other functions (like numpy.fft.rfft()) which work by default on axis=-1.
This works smoothly with always_2d=False, but it fails silently (just giving a completely wrong result) with always_2d=True.
For multi-channel files, people will have to explicitly use axis=0 anyway, there is no way around this (and if they forget, they will get unexpected - a.k.a. wrong - results).

I would expect to be able to do the simple steps

1. load a mono file
2. compute its FFT

without having to specify some cryptic argument always_2d=False.

For people working mainly with mono signals (which is probably the majority of people working in audio and especially speech signal processing), it would be much more natural to have always_2d=False as default setting.

It would also make things simpler for people learning signal processing with NumPy/SciPy.
Here are 2 examples where I had to explicitly add always_2d=False:
http://nbviewer.jupyter.org/github/mgeier/jupyter-presentation/blob/master/jupyter-presentation.ipynb
http://nbviewer.jupyter.org/github/spatialaudio/communication-acoustics-exercises/blob/master/ir-solutions.ipynb

This makes the following steps much simpler and easier to understand, but of course the explicit argument always_2d=False confuses the hell out of students and colleagues.

[bastibe]

Sure, why not. I don't feel particularly strongly about this either way.

[mgeier]

I added a commit (9b2556e) with a possible implementation.
Is it OK?

[bastibe]

Looks good, merge when you're ready.