feat(governance): add automated model card generator

[obielin]

Summary

• Adds src/climatevision/governance/model_card.py — builds Mitchell-style model cards from training config + evaluation metrics, with an optional fairness report attached.
• Renders to both Markdown (for release notes / model registry) and JSON (for downstream tooling).
• Ships a CLI entrypoint at scripts/generate_model_card.py so the release CI pipeline can attach a card to every model version it publishes.
• Required metric set (iou, f1, precision, recall) is enforced — releases without all four fail loudly.

Why

Sprint deliverable: "Create model card generator (automated from training config + evaluation metrics)." Feeds into the governance CI gate and the LLM impact reporter.

Test plan

• pytest tests/test_model_card.py — 6/6 pass
  • card uses config values for name/version/analysis_type
  • missing required metric raises ValueError
  • rendered Markdown includes every required section
  • write_model_card emits both .md and .json sidecars with matching content
  • end-to-end generate(config_path, metrics_path, ...) loads files from disk

Notes for reviewers

• Branched off develop.
• Default content for intended_use, out_of_scope_uses, limitations, ethical_considerations is included so cards have a sensible baseline even when callers do not customise them. Override per-release as needed.

[franchaise]

Required-metric set (iou/f1/precision/recall) is the right floor and ValueError is correctly raised on missing keys — that means scripts/generate_model_card.py will fail loud in CI rather than silently emitting an incomplete card. Markdown + JSON sidecar covers both human and machine consumers.

The default intended-use / out-of-scope / ethics blocks are sensible defaults; we'll override per-release as we tune ecosystem-specific carbon constants. Approving.