Add mlops-tensorflow-starter template

[ulises-jeremias]

Problem

The proposal includes TensorFlow variants for tabular, sequence, image, and distributed
workloads. CPA should avoid three separate TensorFlow templates per data type.

Solution

Create one templates/mlops-tensorflow-starter after the MLOps common contract and
sklearn MVP are accepted.

Scope:

• Type: mlops-tensorflow.
• CPU-first Keras baseline with a tiny fixture/synthetic dataset.
• tf.data input pipeline pattern.
• Training/evaluation with deterministic smoke tests.
• MLflow local tracking consistent with the common contract.
• Export path using SavedModel where appropriate.
• Full CPA template docs and tests.

Constraints:

• No mandatory GPU dependency.
• No large downloads or network calls in tests.
• No GitHub Actions workflows inside the template.
• Data modality examples should be extensions or documented examples, not separate
  base templates.

Acceptance criteria:

• Generated project passes local scaffold validation.
• CI registry scripts know the new mlops-tensorflow type.

[ulises-jeremias]

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