Remove sklearn<=1.6 compatibility code

[solegalli]

Summary

The project already requires scikit-learn>=1.7.0 (pyproject.toml, tox.ini, .circleci/config.yml all already only test 1.7.1), so every if sklearn_version </> parse_version("1.6") branch across the codebase was dead code — the "older sklearn" side could never execute. This PR removes that dead code, keeping only the always-taken >=1.6 branch (the one using check_estimator(expected_failed_checks=...)).

Changes

• feature_engine/tags.py — collapsed the version check in _return_tags(), the shared helper used by ~20 estimator classes.
• 11 tests/**/test_check_estimator_*.py files — collapsed each if/else on the sklearn 1.6 cutoff, dropped now-unused sklearn/parse_version imports.
• tests/test_wrappers/test_sklearn_wrapper.py — also collapsed the _OneHotEncoder() test helper's sparse/sparse_output branch (sklearn<1.2, dead for the same reason). Left the separate KBinsDiscretizer(quantile_method=...) branch (sklearn<1.7) untouched — different threshold, intentionally out of scope here.
• tests/test_creation/test_geo_features.py — __sklearn_tags__ always exists at sklearn>=1.7, so dropped the hasattr() guard.
• tests/check_estimators_with_parametrize_tests.py — deleted entirely. A standalone, non-CI reference file documenting the pre-1.6 parametrize_with_checks() call signature.

Not touched, on purpose: every _more_tags()/__sklearn_tags__() method definition across feature_engine/. _more_tags() is feature_engine's own internal metadata/xfail-checks store (read directly by tests/estimator_checks/*.py), not a legacy sklearn shim — it has no equivalent in the new tags dataclass. __sklearn_tags__() is the current, correct API.

Known pre-existing gap (not introduced here)

tests/test_prediction/test_check_estimator_prediction.py only ever had the dead sklearn<1.6 branch — no >=1.6 version of test_check_estimator_from_sklearn was ever written for the prediction module (its own # TODO already flagged this). Removing the dead branch leaves that module with zero test_check_estimator_from_sklearn coverage. This already existed before this PR (the branch just wasn't reachable); flagging it here as a follow-up rather than expanding scope to write the missing test.

Test plan

• pytest tests/ — identical result before and after this change: 2010 passed, 114 failed (all 114 are pre-existing narwhals-migration WIP failures, unrelated — confirmed by diffing the exact failing test names before/after, not just counts)
• flake8 feature_engine tests — clean
• mypy feature_engine — the one remaining error (datetime/datetime_subtraction.py) is pre-existing and unrelated, confirmed present on the branch before this PR too
• grep -rn 'parse_version("1.6")' across feature_engine/ and tests/ — zero hits