Parameter estimation with nested sampling

[tomasstolker]

• Updated dependencies in requirements.txt.
• Added the run_multinest function in FitModel to use the nested sampling implementation of MultiNest for the parameter estimation. This can be in particular useful to sample multimodal distributions.
• Some changes in add_samples to make it compatible with both the MCMC and nested sampling results.
• Added errors in case the burnin of the MCMC samples is larger than the number of steps.
• Changed the group name of the fit results from results/mcmc to results/fit.
• Improved compatibility with the Exo-REM models, which are now resampled to a constant resolution of 1000 and available here:
  https://people.phys.ethz.ch/~ipa/tstolker/exorem.tgz
  Note that there is an issue with this set of spectra above ~1600 K because not all models properly converged.
• The luminosity is added to a ModelBox if the Teff and radius are available.
• Refactoring of sort_data to reduce its complexity.
• Improvement with the interpolation of missing grid points in write_data when the model spectra are stored in the database. Previously, the missing spectra were interpolated from neighboring Teff points but this is now done with a linear, multidimensional interpolation with griddata from scipy.