Model registry derives wrong algorithm names via `str.lstrip("m_")`

[thomasATbayer]

Model registry derives wrong algorithm names via str.lstrip("m_")

Summary

MotherModelRegistry._load_models() derives an algorithm name from each
m_*.py model module filename using str.lstrip("m_"). str.lstrip removes a
set of leading characters, not a prefix, so any model file whose name starts
with a character that is also present in "m_" gets silently mangled.

The most visible case: a module named m_mlp.py produces the algorithm name
"lp" instead of "mlp", because lstrip strips the leading m, then _,
then the second m as well, stopping only at l.

Location

src/mother/ml/__init__.py, in MotherModelRegistry._load_models():

algo: str = model_file.lower().lstrip("m_")

Why it is wrong

str.lstrip(chars) treats its argument as a set of characters to strip from the
left, repeatedly, until a character not in the set is found. It is not a
prefix removal.

>>> "m_mlp".lstrip("m_")
'lp'          # expected 'mlp'
>>> "m_catboost".lstrip("m_")
'catboost'    # happens to be correct (next char 'c' is not in {'m','_'})
>>> "m_flow".lstrip("m_")
'flow'        # correct by luck

Any module name of the form m_m... (a model family whose name starts with m,
e.g. mlp, mars, mlr) is corrupted. Names that don't start with m after
the prefix are corrupted only if their leading characters fall in {"m", "_"},
which is why the bug went unnoticed for the existing m_catboost, m_lasso,
m_node, m_randomforest, m_tabpfn modules.

Impact

• get_available_algorithms() returns incorrect algorithm tokens (e.g. lp).
• get_model_class_by_algorithm_and_type(algorithm, model_type) cannot resolve
  the mangled name and raises
  ValueError: Unsupported algorithm '<name>' or model type '<type>'.
• Any config-driven or test code that sweeps over get_available_algorithms()
  breaks for the affected model. This surfaced as failures in
  test/unit/test_ml.py once an m_mlp.py module was added.

Reproduction

from mother import ml

# Add any model module named m_mlp.py under src/mother/ml/models/
# containing an AbstractMotherPipeline subclass, then:
print(ml.get_available_algorithms())        # -> [..., 'lp']  (should be 'mlp')
ml.get_model_class_by_algorithm_and_type("mlp", "regression")  # KeyError/ValueError

Fix

Use str.removeprefix (Python 3.9+), which removes the literal prefix once:

algo: str = model_file.lower().removeprefix("m_")

>>> "m_mlp".removeprefix("m_")
'mlp'
>>> "m_catboost".removeprefix("m_")
'catboost'

Suggested follow-ups

• Add a regression test asserting that a module named m_mlp maps to the
  algorithm "mlp" (and generally that algo == model_file with the m_
  prefix removed).
• Audit for other misuses of str.lstrip/str.rstrip where a prefix/suffix
  removal was intended.