fix: import numpy.typing for numpy<2 compatibility

[xyf5432]

Fix #160

Problem

np.typing.XXX is only accessible on numpy < 2.0 if numpy.typing has been explicitly imported earlier.

The README's uv sync install path is safe — uv.lock pins numpy 2.x, so uv users always get a numpy where the annotations resolve. However, pyproject.toml declares numpy with no version constraint, and pip users are not covered by the lock file: a pip install on a machine with numpy 1.x already present (pip keeps existing versions without upgrading) can hit:

AttributeError: module 'numpy' has no attribute 'typing'

when the annotations in this module are evaluated at import time. Declaring a bare dependency without a floor also leaves future lock re-resolutions exposed — a change in the dependency graph that resolves numpy < 2 would silently break the import.

Fix

Add an explicit import numpy.typing before the first use. This injects typing into numpy.__dict__, making all subsequent np.typing.XXX accesses work on every supported numpy version — no reliance on import order or upstream dependency side effects.

  import numpy as np
+ import numpy.typing
  import torch

Verification

• Verified the failure mechanism in isolation on numpy 1.26.4: np.typing.ArrayLike raises AttributeError: module 'numpy' has no attribute 'typing' unless numpy.typing has been imported first; with the import added, the same expression resolves successfully.
• The modified file passes a syntax check.
• The change only adds an import — no behavioral change for environments where the annotation already resolved (numpy >= 2, or when an upstream dependency already imported numpy.typing).