Track: MLOps extensions and capability packs

[ulises-jeremias]

Problem

Gaston proposed CI/CD, distributed training, and data modality variants as template
features. In CPA, these should mostly be extensions layered onto MLOps framework
templates.

Solution

Create an MLOps extensions track.

Candidate extension groups:

• all-mlops-github-actions: CI, Continuous Training, model quality gate, and deploy
  workflows.
• all-mlops-tabular-data, all-mlops-sequence-data, all-mlops-image-data: data
  modality packs if they can be framework-agnostic.
• mlops-sklearn-distributed: Dask/Ray support.
• mlops-pytorch-distributed: DDP/torchrun support.
• mlops-tensorflow-distributed: tf.distribute support.

Track constraints:

• Extensions use template/ and bank-facing root README.md.
• Extensions include generated docs and append to docs/README.md.
• Extensions use partial pyproject.toml overlays.
• Extensions declare incompatibleWith when they write conflicting paths.
• No extension should silently require another extension unless that limitation is
  documented and tested through curated profiles.

[ulises-jeremias]

This item is part of the MLOps roadmap. If you'd like to work on it, comment here.