Add version-to-version data quality and drift checks

[developer-rpai]

What this adds

Pandera schemas (per-dataset, in catalog repos) validate a single version's
shape. Nothing in cfa.dataops checks that a new version still looks like
the dataset it claims to be. This PR adds a reusable
cfa.dataops.quality module for version-to-version checks:

• freshness — newest record within max_age_days of the reference date
• row_count — new version non-empty and within max_change_ratio of old
• columns — no unexpected added/missing columns
• dtypes — shared columns kept the same dtype
• null_rates — per-column null fractions stable within max_delta
• numeric_drift — column means stable within max_std_shift old-std units
• categorical_values — no new/vanished categories in string columns

Results aggregate into a QualityReport (.passed, .failed, .summary(),
.as_dict(), .raise_on_failure() for CI gating), plus a
python -m cfa.dataops.quality --new/--old/--config CLI that exits non-zero
on failure. Docs in docs/data_quality.md, changelog entry included.

Relevant to #148 (Quality Assurance and Testing epic).

How it was tested

• 47 new tests in tests/test_quality.py, including doctests (repo runs
  --doctest-modules on the package): 47 passed
• ruff check and ruff format --check: clean
• Full collectible suite (test_quality, test_soda, test_utils_*):
  99 passed. Note: several existing test modules cannot be collected in
  this sandbox (missing mako/pydantic/rich/tomli); that is a
  pre-existing environment limitation, unrelated to this change.
• The new module imports only pandas + stdlib, so it adds no new
  dependencies.

Open questions for reviewers

• Is cfa.dataops.quality the right home, or would you prefer this under a
  different module name?
• Worth adding polars-native paths later, or is pandas sufficient given the
  Pandera-based validation in catalog repos is pandas-based?

Draft PR — happy to rework based on feedback before review.