fix(transformer): validate dataframe schemas

[alisadeghiaghili]

Summary

• Reject duplicate pandas column labels before backend-specific validation in transformer fit and transform.
• Specify pandas-DataFrame-only container support with actionable TypeError contracts for NumPy arrays and SciPy sparse matrices.
• Preserve caller data on rejected inputs and document the public schema boundary.

Validation

• pytest -q -p no:cacheprovider tests/test_transformer.py (55 passed)

uff check missingly tests

• mypy missingly/transformer.py --follow-imports=skip --ignore-missing-imports (0 errors)
• Installed-module transformer doctest (32 attempted, 0 failed)
• git diff --check and changed-lines <=88

Coverage and risk

Baseline is the official Python 3.12 artifact on main: 92.56% (3806/4112), 306 missed. This PR does not change any coverage configuration, threshold, omission, or exclusion; CI artifact will determine the delta. The behavioral change is intentionally fail-fast: duplicate labels now raise stable ValueError messages instead of leaking backend-specific errors. NumPy and SciPy sparse inputs remain unsupported and are now contract-tested; optional Polars/Arrow runtime paths were not exercised because this estimator publicly accepts pandas DataFrames only.

Refs #78
Refs #60
Refs #69