Merge pytorch, tf2, and carla mot images

[christopherwoodall]

• [x] Consolidates Armory's pytorch, tf2, and carla images
• [x] Remove old images
• [x] Add new image to CI
• [x] Update hash map
• [x] Update docker/build.py
• [x] Update dependencies in pyproject.toml and environment.yaml
• [x] Bump minimum Python version to 3.8
• [x] Update documentation

[davidslater]

If we are going to move in this direction, let's also remove tf2 and pytorch-deepspeech, and just move directly to one container.

[davidslater]

TF2 can be mapped to PYTORCH. Actually, I would probably just map them all to armory or some generic container name.

the deep speech container will require some additional work: a) removal of armory/baseline_models/pytorch/deep_speech.py b) pinning of eval1-4/asr_librispeech/, eval5/asr_librispeech/ to 0.16.1 version docker container (and adding deprecation) c) expanding eval6/asr_librispeech to include targeted and targeted entailment attacks d) various updates of docs

[christopherwoodall]

Added fix for #1787

[christopherwoodall]

I agree with the above, but I think we should scope that work out into a different PR.

[davidslater]

Let's have pytorch-deepspeech be a different PR, but let's put tf2 in this one, mapping it to PyTorch like you have done here with carla-mot. The current tf2 and pytorch docker images are effectively identical.

[christopherwoodall]

Split PR. See PR #1807