Bounds given as a list or tuple raise AttributeError when params is a numpy array #725

Bug description

When params is a 1d numpy array and no constraints or scaling are used, get_converter takes a fast path (_get_fast_path_converter in src/optimagic/parameters/conversion.py). That fast path calls bounds.lower.astype(float) and bounds.upper.astype(float) directly instead of going through get_internal_bounds. As a result, om.Bounds with a list or tuple as lower or upper fails with AttributeError. The deprecated lower_bounds=[...] argument fails the same way, because it is converted into Bounds(lower=list).

The same bounds work as soon as the problem leaves the fast path, for example with scaling=True, because the slow path processes bounds through get_internal_bounds. The fast path also skips the lower > upper check and the conversion of all-infinite bounds to None that get_internal_bounds performs.

To Reproduce

import numpy as np
import optimagic as om

params = np.array([1.0, 2.0, 3.0])

om.minimize(lambda x: x @ x, params, algorithm="scipy_lbfgsb",
            bounds=om.Bounds(lower=[0.5, 0.5, 0.5]))
# AttributeError: 'list' object has no attribute 'astype'

om.minimize(lambda x: x @ x, params, algorithm="scipy_lbfgsb",
            bounds=om.Bounds(lower=[0.5, 0.5, 0.5]), scaling=True)
# works, params = [0.5, 0.5, 0.5]

Expected behavior

Bounds that work on the slow path should also work on the fast path. Calling get_internal_bounds(params=params, bounds=bounds) in _get_fast_path_converter would do this. It already has its own fast path for 1d array bounds, so the behavior for array bounds stays the same. I have a small patch with a regression test in tests/optimagic/parameters/test_conversion.py and can open a PR if that is welcome.

System

• OS: macOS 15
• Version: optimagic main at 2b1eea8 (0.5.5.dev17), Python 3.12

Disclosure: found, reproduced and the draft fix tested locally with the help of an AI coding agent (Claude Code); reviewed before filing.