tests: plan a better test suite structure and naming scheme

[janfb]

At the moment we have some structure in the test suite, e.g., roughly a test module for every method or feature, e.g., mcmc_test.py, or embedding_net_test.py.

However, we also have things like linearGaussian_snpe_test.py (and the same for many inference methods). We should think about a more organized and systematic naming scheme and testing strategy in general:

• consistent naming
• make testing of inference methods more modular: e.g., by separating tests for density estimators and classifiers from tests for inference methods like NLE, NRE, NPE). We then have to test the estimator accuracies only once, and can then use fixtures and more functional tests for the specific inference methods NPE, NLE, NRE etc. This will reduce repeated training of estimators in the test suite.
• more ideas?

Note, this issue is more about making a plan how to refactor the test suite in general. The concrete implementation we will need to discuss and coordinate with other issues.