Calibr8

+ The `plot_model` function was polished to produce nicer looking plots by default:
+ The subplots are now created with a fine tuned layout that is (almost) square and has nice looking margins inbetween the plots.
+ Log-scale is automatically turned off when calibration data is <= 0.
+ Residual ylimits are now symmetric.
+ A `band_xlim` kwarg was added to configure the range of the likelihood bands in the left two plots.

+ Fixes a severe performance degression of the `infer_independent` method by vectorizing an evaluation of the `loglikelihood`.

+ `infer_independent` was refactored: It is now implemented in `CalibrationModel` and calculates both ETI and HDI by default.
+ New method `CalibrationModel.likelihood` that can be parametrized with `scan_x=True` to evaluate the likelihood at all entries in `x`, instead of doing elementwise evaluation along `x` and `y`.
+ Imports work with `theano-pymc v1.1.2` (used by PyMC3 `v3.11.1`) and `aesara v2`
+ Added option to plot relative residuals in `plot_model` and `plot_t_band`
+ Guessing functions were removed because they are not very useful in practice
+ License document (and setup.py) were updated.

+ Terminology was changed `ErrorModel` becomes `CalibrationModel`
+ Fitting functions automatically ignore inf/nan
+ `NumericPosterior` has new properties `hdi_lower` and `hdi_upper` that give direct access to the lower and upper ends of the HDI

Release for Zenodo