release: 0.6.0

[andremsouza]

Version bump and changelog for 0.6.0. Nothing breaks, no results change.

What 0.6.0 is

Two things, both additive:

• A map can be used as a scikit-learn estimator. fit, transform, predict, score,
  get_params, set_params on SOM, plus python_som.sklearn.SOMEstimator for full Pipeline and
  GridSearchCV integration behind pip install "python-som[sklearn]".
• The plain-string deprecation from 0.5.0 is withdrawn. Strings are permanent.

train is untouched and remains primary. numpy is still the only runtime dependency.

The withdrawal, stated plainly

0.5.0 shipped a DeprecationWarning announcing that mode="batch" would stop working in 1.0.0. That
was wrong, and this release reverses it. Every comparable library passes options as plain strings and
none export enums — scikit-learn, numpy, scipy, minisom, sompy — so removing the string form would have
made this the only library in its ecosystem to reject mode="batch".

Anyone who migrated to the enums while 0.5.0 was current keeps working code: the enums are staying too.
The changelog says so directly rather than quietly dropping the warning.

Verified from the built wheel, both ways

numpy only:
  version 0.6.0
  estimator methods: predict (40,)  transform (40, 30)  score -0.5236
  plain strings under -W error::DeprecationWarning: silent
  python_som.sklearn: correctly unavailable, with a message naming the extra

with [sklearn]:
  GridSearchCV(SOMEstimator(y=4), {"x": [4, 6]}, cv=3) -> best {'x': 6}, score -1.1440

Plus 424 tests at 100% coverage, ruff / ruff format / mypy --strict clean, mkdocs build --strict
exits 0, twine check passes both artifacts.

Where 1.0.0 stands

Deliberately not next. What remains breaking is thin — keyword-only arguments, removing two setters,
removing pre-0.3.0 private aliases, dropping Python 3.10 after its October 2026 EOL — and the estimator
interface introduced here should have real use before the API is frozen around it. refactor/enums-only
is dropped from the plan entirely.

After merge

Tag v0.6.0, then the workflow builds and waits on the pypi environment reviewer, which you
authorise in the GitHub UI.