Isochrones

This release both makes MIST version 1.1 default (notably including the new Gaia G, BP, and RP bands, as well as the TESS bandpass) and simplifies how `isochrones` downloads model grids. Before, it had downloaded pre-packaged grids for both Dartmouth and MIST that were gzipped tar files of selected photometric systems. Now, individual tarballs are downloaded as those bands are requested. And if you really want to use MIST 1.0 still, you may still initialize it with a `version=1.0` argument. You may also try out a the MIST grids with v/vcrit = 0.4 (rather than the default non-rotating v/vcrit = 0) by initializing your MIST isochrone with the `vvcrit=0.4`; e.g. `mist = get_ichrone('mist', vvcrit=0.4)`.

When you upgrade to this version, I recommend that you **remove** your current `~/.isochrones` (or `$ISOCHRONES`) directory so you can start fresh and clean. This will clear out all the pre-calculated/formatted HDF grids, so you can be sure of actually using MIST 1.1 instead of reading in the old HDF files.

Changes/bugfixes

* Makes sure HDF files do not remain open after `StarModel.load_hdf`.
* Adds parallax (if present) to the information displayed with `StarModel.print_ascii`.
* Adds support of pickling `StarModel` objects.
* Corrects bug with `Isochrone` being called with different length arguments.
* Includes correct test data to allow unit tests to run.
* Adds Gaia DR2 query support.

Known issues
* Documentation is not yet fully updated.
* Treatment of extinction is still not appropriate for very large bandpasses. **Do not fit to Gaia G band.**
* Isochrone interpolation is still unreliable for evolved stars, especially with the MIST grid interpolation as currently implemented. **Be suspicious if fitting subgiants or giants.**

This version will now be compatible with [vespa](http://github.com/timothydmorton/vespa), after vespa's next upgrade.

Also this release fixes a major issue with the `dartmouth.tri` file that was breaking with scipy 0.19. Everything should also now be python 3-compatible.

To use 1.1, you will have to delete the dartmouth tables in your `.isochrones` (or `$ISOCHRONES`) directory, and then let the code re-download them for you.

Documentation has still not been updated, but will be with the next release. This is a functional quick-fix for the major issues, and documentation will follow.

First release of version 1.0. Significant changes from previous versions.
- MIST models are now available (much broader coverage in mass, age, and metallicity), in addition to Dartmouth.
- Book-keeping is much more organized with the new `ModelGrid` object, which takes the model grids exactly as published and does all the parsing as part of the package. This should make it easier to extend to using other model grids.
- `BinaryStarModel` and `TripleStarModel` no longer exist; now you can just provide an `N=2` keyword argument to `StarModel`.
- Support for non-trivial collection of photometric bands for multiple stars: e.g., pair of stars unresolved in 2MASS but resolved in high-resolution imaging.
- `Isochrone` object can now predict `density`, `nu_max`, and `delta_nu`, to help support transit host stars and asteroseismic stars, where such data is available.

Big changes to how `StarModel` fits happen now---fitting now defaults to using MultiNest via PyMultiNest. To take advantage of this change, you must install MultiNest and PyMultiNest. If you don't have these packages, you can follow [these directions](http://astrobetter.com/wiki/MultiNest+Installation+Notes) to install them.

MultiNest does a much better job than `emcee` of exploring the stellar model parameter space when only photometry is being used as observational constraints, because the posterior can be quite multi-modal. As a result, you must use MultiNest if you want a fair fit when you're just using photometry.

The way to fit a `StarModel` now is using the `.fit()` method, which will default to `.fit_multinest()` if MultiNest is available, and will otherwise default back to using `emcee` via `.fit_mcmc()`. Please let me know if you have any troubles with this.

The `starfit` command-line script should work the same as before; again defaulting to MultiNest if available.

In addition, I have finally included unit tests: type `nosetests isochrones` on the command line to run them. If all goes well, everything _should_ be OK.

Another note---- I have also fixed a bug in the dartmouth triangulation file used for interpolation; if you have been using `isochrones` before with the Dartmouth grids, you should get an error now when you try to load them, asking you to delete and re-download the `dartmouth.tri` file. Again, let me know if you have a problem with this.

Support has now been added to include parallax (in miliarcsec) to the observable properties of a `StarModel`.

There are now also two command-line scripts, `starfit`, which allows executing an MCMC fit of a `StarModel` from the command line (e.g. `starfit mystar` where `mystar` is a folder containing a `star.ini` file), and `batch_starfit`, which executes many starfits at once on a cluster (currently assuming slurm is the batch scheduler, but presumably could be adjusted to any other).

Also, there was a correction made to the `loglike` method of `StarModel` such that now I believe the likelihoods/priors are properly normalized; this was not the case before.