[akifkulll]

Closes #5

Summary

Adds Pipeline, a BasePreprocessor that runs blocks in order and feeds each block's output into the next.

• Steps are blocks or (name, block) pairs. Unnamed blocks are named after their lowercase class name; repeated class names get -1, -2 suffixes.
• fit_transform runs the chain once and reuses each step's output (instead of fit().transform(), which would run every step twice and would be wrong for blocks whose fit_transform differs, such as a future cross-fitted target encoder).
• Nested get_params / set_params (<name>__<param>), step replacement by name, named_steps, and get_feature_names_out chained through every step.
• Step blocks are fitted in place, as in scikit-learn.

Design decisions beyond the issue text:

• Step names are validated wherever they are needed (fit, get_params, set_params, named_steps), not only in fit: GridSearchCV calls get_params before fitting, and duplicate names would otherwise silently merge two steps' parameters. __init__ still does no validation.
• "steps" is reserved as a step name (it would collide with the steps parameter in get_params).
• An explicit name that collides with an automatic one raises ValueError instead of renumbering, so adding a step never changes another step's address.
• A replaced unnamed step is stored as (name, block), so its name and every <name>__<param> address stay the same. The user's list is never modified.
• set_params validates the whole call first, recursively through nested pipelines; an invalid call changes nothing.
• An empty pipeline raises ValueError.

How was it tested

• tests/test_pipeline.py: 37 tests with small test blocks defined in the file (one renames columns, to cover get_feature_names_out chaining).
• ruff format --check ., ruff check . and pytest pass on all three CI configurations: Python 3.10 + scikit-learn 1.7.2 (106 passed), Python 3.13 + scikit-learn 1.9.1 (106 passed), Python 3.10 + scikit-learn 1.3.2 (105 passed, 1 skipped: tags API). Also 106 passed locally on Windows, Python 3.12.
• Manual check with the real blocks: Pipeline([("imputer", MissingValueImputer()), FeatureScaler()]) inside make_pipeline(..., SVR()) and GridSearchCV over pipeline__imputer__strategy and pipeline__featurescaler__method.

Checklist

• ruff format . and ruff check . pass
• pytest passes
• New or changed blocks follow the block rules in CONTRIBUTING.md
• Tests include a leakage test (statistics come from the training data only) and a run inside sklearn cross_val_score