Pygpso

The new version brings a few changes:
* switch from Travis to github actions for CI (linting, testing, and codecov)
* fix saving of `GPSurrogates`

New minor release:
* only `GPFlow` in requirements, so that it installs the best version of `tensorflow` and `tensorflow-probability`
* better testing scheme for saving / loading of `GPRSurrogate`

* easier interface for surrogate GP creation and control
* more surrogate options
* python3.8 support
* wider testing suite (testing regularly 3.6 - 3.8)
* DOI via zenodo

- switch from GPFlow2.0rc to official 2.0 release now on `pypi`!
- that means that `pygpso` can be installed via `pypi` as well

New release comes with new features!

* saving of optimiser state: now possible to resume optimisation from the saved state
* saving of GPR model
* saver helper for saving modelling output, not just scores which we are optimising for
* support for callbacks - custom functions run repeatedly in the particular state of the optimisation
* support for repeated evaluation of the objective function for the same parameters (useful for a stochastic systems optimisation - the scores are then aggregated to be represented by a scaler per parameters)
* better example notebooks (also support of interactive run thanks to binder)
* couple minor bug fixes

- `igraph` as an optional dependency