Add MLOps data modality extension packs

[ulises-jeremias]

Problem

The proposal multiplies templates by data modality: tabular, sequence, and images. This
creates nine non-distributed templates before adding distributed variants.

CPA should avoid this explosion if modality-specific code can be layered as examples or
extensions.

Solution

Design data modality packs as framework-agnostic extensions where possible.

Candidate extensions:

• all-mlops-tabular-data
• all-mlops-sequence-data
• all-mlops-image-data

Scope:

• Add small local fixtures or synthetic data generators.
• Add data validation/preprocessing utilities that do not depend on a specific model
  framework when possible.
• Add tests for schema/shape/leakage rules.
• Add docs:
  • template/docs/TABULAR_DATA_GUIDE.md, or equivalent;
  • template/docs/SEQUENCE_DATA_GUIDE.md;
  • template/docs/IMAGE_DATA_GUIDE.md;
  • template/docs/README.md.append.

Constraints:

• Do not write framework-specific model/training files in all-* modality packs.
• If a modality requires framework-specific code, split it into framework-specific
  extensions such as mlops-pytorch-image-data rather than forcing generic paths.
• No large sample datasets or network downloads.

Acceptance criteria:

• The first modality pack composes with mlops-sklearn-starter and has a CI profile.
• The design avoids duplicating the same data code across framework templates.

[ulises-jeremias]

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