Sedfitter

What's Changed

- Small astropy 4.0 compatibility refactor by keflavich in https://github.com/astrofrog/sedfitter/pull/67
- Update package infrastructure and fix compatibility with latest astropy versions by astrofrog in https://github.com/astrofrog/sedfitter/pull/79
- ndistances can't be a float in recent versions of numpy by keflavich in https://github.com/astrofrog/sedfitter/pull/73
- Use memmap when reading in model grids by keflavich in https://github.com/astrofrog/sedfitter/pull/78
- More infrastructure improvements by astrofrog in https://github.com/astrofrog/sedfitter/pull/80

**Full Changelog**: https://github.com/astrofrog/sedfitter/compare/v1.3...v1.4

- Fixed an issue that occurred during the convolution if not all SEDs
- had the same number of wavelengths. [66]

- Use install_requires instead of requires.

- Fix color of lines in SED plots. [52]
- Fix Windows compatibility. [53]
- Fix compatibility with Astropy 2.x and above. [60]

- Fixed a bug in the calculation of the reduced chi^2: this should
- have used the number of valid data points (valid flag 1 or 4) but
- instead was using the total number of valid and invalid points.
- [49]
- Fixed compatibility with Python 3.6. [49]

- A valid flag of '9' can now be used to denote a flux that should
- be plotted but not fit. [32]
- Fix issues that occurred if no valid fits were present for a given
- source. [38]
- Switch the order of the dimensions in SED flux cubes to optimize
- performance. This means that this version of the SED fitter will be
- incompatible with flux cubes generated with earlier versions, but
- will result in significant performance improvements when plotting
- SEDs. [44]
- Add a `memmap` option to `convolve_model_dir` that controls whether
- memory mapping is used to read the flux cubes (in the case where
- SEDs are stored in cubes rather than individual files). If the cubes
- can fit in memory, the convolution is much faster if the memory
- mapping is explicitly turned off.