- Completed a variable name refactoring (from vec_level to emb_level) across multiple files.

…beddings object

- Now the anonymization pipeline is completely separated from the anonymizer object.
- Specify a BaseAnonymizer subclass in a config, then import that using !include syntax. Example is shown for GAN anonymizer.
- Add a Passthrough anonymizer

- Can convert the .scp files to absolute or relative.

- Makefile installs the environment local to the project folder.

- each libespeak backend instantiation creates a copy of the library.
- for some reason previous instances (for each utterance) are not garbage collected.
- latest at 1500 or 2000 times of ProsodyExtraction the anonymization pipeline crashes.
- As a temporary solution, since ProsodyExtraction does not support `n_processes>1`, we can make different instances share the backend, as long as specs do not change. This also accelerates the prosody extraction.

A better fix could be performed to also allow parallel execution but I couldn't see how today.

…mization and evaluation pipelines

- Also fixed a duplicate `save_intermediate` entry in `anon_ims_sttts_pc.yaml`

- BaseAnonymizer and its descendants are now dumpable
- Improved documentation
- The configs now use !PLACEHOLDER tag from HyperPyYAML instead of the custom class

…ion too

- Use absolute paths for `asr.sh` invocation
- Improved documentation

- Add missing !PLACEHOLDER tags
- Define two dataset configs, one for anonymization, one for evaluation (vctk all/common/diff requires separate treatment)