[ChrisW09]

Fixes #535

Problem

set_seed() validated seeds with isinstance(seed, int), so NumPy integer scalars raised ValueError: seed must be a non-negative integer. These come from rng.integers(...), np.arange(...) or a pandas column. The check failed both for direct calls and for random_state=np.int64(...) on any estimator. TabularDataModule also passed random_state straight to torch.Generator.manual_seed(), which needs a Python int.

Fix

• deeptab/core/reproducibility.py: set_seed() accepts any non-negative numbers.Integral and converts it with int() before seeding Python, NumPy and torch. Non-integers (1.0, "7", np.float64) are still rejected. The numpydoc mentions NumPy integer scalars.
• deeptab/data/datamodule.py: TabularDataModule stores random_state as a Python int (or None). Python and NumPy integers are converted, and the parameter is annotated int | np.integer | None; floats and strings raise TypeError instead of being truncated by int().

Tests

In tests/core/test_random_state.py:

• TestSetSeedPrimitives::test_numpy_integer_seed_matches_python_int[int64|uint32|int32]: the same torch draws as the equivalent Python int.
• TestSetSeedPrimitives::test_non_integer_seed_is_still_rejected[float|str|float64]
• test_datamodule_rejects_non_integer_random_state[float|str] and test_datamodule_converts_numpy_integer_random_state
• test_estimator_accepts_numpy_integer_random_state: an MLPRegressor with random_state=np.int64(...) gives the same predictions as with the Python int.

The NumPy-integer tests fail on main and pass with this change.

Docs

The set_seed docstring and the TabularDataModule random_state parameter are updated. No page changes needed.

Checks

• just test: 3246 passed, 46 skipped, 28 xfailed; coverage 91.29%
• just docs: build succeeded (warnings treated as errors)
• just check: all hooks passed (ruff, prettier, pyright)