[ENH] add _normalize_dist_str utility for consistent distribution string aliases

[paramsureliya]

Part of #1023

Reference Issues/PRs

Part of #1023 first step toward uniformizing distribution string handling across all probabilistic regressors.

What does this implement/fix? Explain your changes.

Adds a shared _normalize_dist_str utility (skpro/regression/_dist_utils.py) that maps case-insensitive distribution string aliases to the canonical capitalized class name used internally in skpro, e.g. "gaussian" → "Normal", "lognormal" → "LogNormal", "t" → "TDistribution".

Integrates this into NGBoostAdapter (inherited by both NGBoostRegressor and NGBoostSurvival) as a proof of concept. Normalization happens at point-of-use inside the adapter methods, not in __init__, so get_params()/clone() compatibility is preserved.

The remaining regressors from #1023 (XGBoostLSS, ResidualDouble, CyclicBoosting, GAMRegressor, GLMRegressor, GlumRegressor) will be addressed in follow-up PRs.

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

No new dependencies.

What should a reviewer concentrate their feedback on?

• The alias map in _dist_utils.py any missing or incorrectly mapped aliases?
• The integration pattern in NGBoostAdapter normalization into a local dist variable at point-of-use, leaving self.dist unchanged for sklearn compatibility.

Did you add any tests for the change?

Yes skpro/regression/tests/test_dist_utils.py with 57 tests across 3 classes:

• TestNormalizeDistStr: unit tests for every alias, canonical passthrough, non-string passthrough, unknown string warning, and idempotency
• TestCrossRegressorAliasConsistency: invariant tests ensuring the same alias resolves identically regardless of which regressor calls the function
• TestNGBoostRegressorAliases: end-to-end integration tests for NGBoostRegressor (auto-skipped if ngboost not installed)

Any other comments?

None.

[fkiraly]

This is great!

If it is already done though, we should apply it to all places where distribution (string) aliasing happens, consistently.

Further, I would only advertise the "canonical" string in the docstrings, in order to uniformize future usage. As "canonical" strings, I would use the class names as in skpro. Alternative aliases for downwards compatibility can still function but be unadvertised.

[paramsureliya]

Hi @fkiraly I've applied _normalize_dist_str to all remaining regressors. can you please review it.

[fkiraly]

Thanks!

Question, why are you reformatting or changing docstrings that looked like they aere good? E.g., why are you changing NGBoostRegressor, and why are you introducing newlines in places unrelated? Looks like AI use.

[paramsureliya]

Hi @fkiraly fixed the isort failure.
Regarding the NGBoostRegressor docstring change: I updated it based on your earlier feedback to "only advertise the canonical string in the docstrings." Happy to revert it if you prefer to keep the original format there just let me know.