Wrapper for blockmedian

[weiji14]

Description of the desired feature

Implement blockmedian that 'Block averages (x,y,z) data tables by L1 norm". This is a data filtering function that I've used in my work before, and one I'd like to incorporate into the upcoming PyGMT FOSS4G workshop #317 (if things move quickly enough!).

The implementation will follow in the footsteps of the surface function #243, using the same @tut_ship.xyz example dataset for the unit tests. I'm looking to implement blockmedian under a new file called filtering.py which will eventually hold other GMT filtering functions like blockmean and blockmode.

Currently I'm facing a problem with deciding how to handle the inputs/outputs. There are various combinations, and I'm thinking of using a 'what you put in is what you get out' strategy, I.e. :

Output\Input	file	pandas table	numpy array*	x, y, z triples**
file	x
pandas table		x
numpy array			x
x, y, z triples				x

* numpy array could also be a python list of list
** could also be x, y, z, w quadruplets

A tempting alternative is to only provide a pandas.DataFrame table as an output, and if a pandas.DataFrame was provided as input, we make sure to copy the column names to the output table as well. This closely follows the grdtrack example at #308.

Thoughts?

Are you willing to help implement and maintain this feature? Yes