Fastmat

- Introduce pickle support for fastmat matrices
- Introduce slicing syntax to retrieve rows, columns and elements
- Introduce new documentation scheme using Sphinx (http://fastmat.readthedocs.io)
- Introduce NDFourier matrix class
- Introduce ARM support (tested on ZYNQ SoC-platform)
- Introduce inspection routines
- Improve CG algorithm
- Rework matrix properties, data types and performance profiling
- Rework operator overloading
- Major and Minor bugfixes in BlockDiag, Circulant, CG, Hadamard, Product, Toeplitz, SAFT Demos and normalization routines

- Introduce new classes: DiagBlocks, MLCirculant, MLToeplitz, MLUltraSound
- Introduce new algorithm: FISTA
- Introduce new API functions: internal memory strides for efficient slicing operations
- Improvements in Fourier, Hadamard, LFSRCirculant, Toeplitz,
- Introduce support for Mac OS X
- Largely improved testing, continuous integration and deployment
- Major and Minor bugfixes
- Rework source code, improve readability
- Improve setup.py, fix package dependency locks during install
- Introduce wheels for x86 and amd64 architectures on linux, windows and mac OS X