Pyqt-fit

- Corrected direct setting of bandwidth / covariance
- Added doc for convolution

- Correct use of kernel in non-parametric regression
- Added attributes to BootstrapResult
- Added efficient binning in Cythong and Python
- If using Cython, linear binning is used, which is better for small bin numbers
- Using rfft to save time (computes half the values)

- Removed necessity of calling fit
- Added warning to main text
- Small FFT optimization

- Removed unnecessary type conversion
- Rewrote the non-parametric fitting
- Rewrote parametric fitting
- The interface should be now uniform across the board, but incompatible with previous versions

- Using 'out' rather than 'output' for in-place operations to conform to numpy convention
- Added a kernels.Kernel base-class implementing cdf, pm1, pm2, fft and dct in term of the pdf
- Corrected all the kernels, both Python and Cython
- Added and restructure kernel tests
- Restructured KDE tests.
- Moved TransformKDE to the kde_methods
- Updated tutorial
- Added decorators for 1 argument numpy transformations (i.e. functions working
on arrays and returning an array of same shape as the input)

- Corrected KDE code to pass tests
- All tests pass for Python2 and Python3
- Corrected loader for python 3.4