Semi-major release!

• Change the way fits are stored in the resonator object. This slightly breaks backwards compatibility.
  Results that used to live at Resonator.lmfit_result or Resonator.emcee_result now live at
  Resonator.lmfit_result['default']['result'] and Resonator.emcee_result['default']['result']. This allows the user to store
  multiple fit results for the same resonator by specifying a label for the fit (as opposed to using the
  'default' label).
• Update dependency on matplotlib to be >= 2.0
• Fix minus-sign error in cmplxIQ fit function and params guessing (each had one error, was cancelling)
• Add a burn_flatchain method to each resonator to allow burning off some samples from the mcmc analysis
• Add a function to ESL_tools to read in binary fits files coming out of ESL
• Add a mask option to process_file to allow for masking data via a slice object
• Change under the hood to how fitting works for Resonator objects (and eventually to ResonatorSweep data as well).
  Now instead of being forced into fitting only I/Q data, fit functions take a Resonator object as an argument.
  This gives the fit function author access to all of the attributes and data stored in the object.
• Update the complxIQ fit function to calculate not just the model, but the baseline, or even the model without the baseline.
• Allow the user to pass a frequency vector to cmplxIQ to specify which frequency points to calculate the model at.
• Fix a small bug in plot kwargs checking introduced during update to Python 3.