drop pytorch-deepspeech and tf2 builds

[mwartell]

In gard sync today @davidslater suggested we unify the armory flavors because it appears we can.

Because Armory uses tensorflow-datasets, it requires both that and (transitively) tensorflow2, so it is reasonable to fold those into what is now called the twosixarmory/pytorch image. It was suggested these become something like twosixarmory/armory

This obsoletes the armory/tensorflow and armory/pytorch-deepspeech container builds, and we'll need to revise

• [ ] documentation
• [ ] CI Build
• [ ] CI test
• [ ] scenario_configs which reference the obsolete containers

cribbed from Slack 3-Oct-2022:

I’m considering removing the tf2 and pytorch-deepspeech containers and (most of) their requirements. The former only provides additional support for the TF MSCoco model that is used for DApricot, which we are no longer using. DeepSpeech 2 is pretty old and adds a bunch of weird dependencies. It seems much more straightforward to use TORCHAUDIO.MODELS Conformer or Wav2Letter. We can then keep just transformers and librosa , as they are needed in our codebase anyway.

This would drastically simplify our Docker containers, as we would basically have just 1. (pytorch already installs TFDS and Tensorflow, so we would just use that and probably rename it to armory or armory-core or something like that).

That would also drastically simplify our python dependencies and pip installs.

Going forward, we could effectively reject contributions to our codebase that involves weird dependency management, but instead require people to use the more standard torch / tensorflow versions. If we really needed to support those, we could instead support them as one-off docker containers for a specific scenario.

[mwartell]

As of armory v0.16.0 we have opted to continue providing tensorflow and deepspeech containers and added the carla-mot image flavor.

These are still hoped to be removed by 0.17.0, and the carla-mot has been explicitly flagged, at release, as being "may go away after eval 6".

[davidslater]

For the version 0.17.0 release, we should have deprecation errors that tell users to use version 0.16.0 for those containers.

[christopherwoodall]

See @davidslater comments on migrating the deep speech image here.

[mwartell]

Fixed by #1786