[BUG] use instance-level tags in _check_estimator_deps

[Me-Priyank]

Reference Issues/PRs

Fixes #600. See also sktime/sktime#10520, sktime/sktime#10528.

What does this implement/fix? Explain your changes.

_check_estimator_deps, _check_python_version and _check_env_marker read the python_dependencies, python_version and env_marker tags via get_class_tag, so dynamic tag overrides on instances were ignored.

This PR adds a private helper _get_tag in skbase/utils/dependencies/_dependencies.py, used in all three places:

• for instances, tags are read via get_tag(..., raise_error=False), i.e., including dynamic tag overrides;
• for classes, tags are read via get_class_tag, unchanged. Objects without get_tag also fall back to get_class_tag, as before.

Also:

• updated the obj docstrings of the three functions accordingly, and corrected the obj description of _check_env_marker (previously "used to check python version");
• registered _get_tag in SKBASE_FUNCTIONS_BY_MODULE in skbase/tests/conftest.py;
• added test_check_estimator_deps_dynamic_tags, parametrized over the three tags. It checks both directions: a failing class tag cleared on the instance passes, and a failing tag set only on the instance fails. The test fails on main and passes with this change.

Does your contribution introduce a new dependency? If yes, which one?

No.

What should a reviewer concentrate their feedback on?

• the helper semantics: instances use get_tag, classes and objects without get_tag use get_class_tag
• downstream effect on sktime once it allows a release containing this:
  • ignore_deps=True works as documented, e.g., CiscoTSMForecaster with its own test parameters now fits and predicts with its dummy model without cisco-tsm installed;
  • in environments without the soft dependencies, 4 tests in sktime/forecasting/tests/test_dynamic_dep_tags.py fail (ChronosForecaster, Chronos2Forecaster), because their __post_init__ imports transformers/torch. Today __post_init__ is only skipped because the dependency check fails. sktime currently pins scikit-base<1.2.0, so nothing breaks until that bound is raised; the sktime side likely needs coordinating with [ENH] Isolate torch imports in forecasting module sktime#11087 and [ENH] Introduce BaseFoundationForecaster to unify zero-shot foundation models sktime#10508.

Any other comments?

Local checks: full skbase test suite incl. doctests (1620 passed, 23 skipped), skbase/_nopytest_tests.py, and all pre-commit hooks on the changed files pass.

PR checklist

For all contributions

• I've reviewed the project documentation on contributing
• I've added myself to the list of contributors.
• The PR title starts with either [ENH], [CI/CD], [MNT], [DOC], or [BUG] indicating whether
  the PR topic is related to enhancement, CI/CD, maintenance, documentation, or a bug.

For code contributions

• Unit tests have been added covering code functionality
• Appropriate docstrings have been added (see documentation standards)

[Me-Priyank]

@fkiraly PTAL