fix: NumPy 2.4+ compatibility in feature selection fitting paths

[KenyaOtsuka]

Summary

Fix NumPy 2.4+ compatibility issues in the feature-selection fitting paths of src/fastl2lir/fastl2lir.py.

Background

In the feature-selection paths of __sub_fit and __sub_fit_save_select_feat, the right-hand side passed to np.linalg.solve() was reshaped to a column vector using .reshape(-1, 1).

As a result, the solution Wb also became a 2D column vector. Expressions such as Wb[i] therefore returned a length-1 array rather than a scalar. Assigning such a length-1 array to a scalar element of W or b was deprecated in NumPy 1.25 and raises a TypeError in NumPy 2.4+.

Changes

• In __sub_fit, update both feature-selection branches to pass a 1D RHS to np.linalg.solve().

• Replace the per-feature assignment loop with a vectorized assignment:

  W[index_outputDim, I[:-1]] = Wb[:-1]

• In __sub_fit_save_select_feat, apply the same 1D-RHS pattern and replace the assignment loop with:

  W[index_outputDim, I] = Wb[:-1]

• Replace dtype=np.bool with dtype=np.bool_ for compatibility with NumPy versions where np.bool was removed.

The no-feature-selection multi-target solve path in __sub_fit is left unchanged, since that path intentionally solves a multi-target system.

Tests

I ran the following tests and confirmed that all tests pass, including with NumPy deprecation warnings treated as errors:

uv run pytest                               # 7 passed
uv run pytest -W error::DeprecationWarning   # 7 passed