Fix consistency labels ordering and simplify

[CeliaBenquet]

• I removed the possibility of using both the between-runs and between-datasets at the same time.
• I removed the possibility to add dataset_ids for the between-runs consistency. The labels will be np.arange(len(embeddings)) always.
• The embeddings provided for between-runs must be the same number of samples (as no alignment is happening), but the number of output dimensions can be different.
• I re-wrote the tests: 2 functions one for each (datasets / runs).
• I re-wrote some of the examples in the usage file so that the tests pass.

to do:
Maybe need better documentation in the usage file for instance.

---

Fix #27
Fix #54

[stes]

lgtm, we can close other PR and use this