Bump minimum numpy to 2.2.6 (#201)

* Bump minimum numpy to 2.2.6

This version solves the build error with Python 3.13 (and the associated failing test) while maintaining compatibility with Python 3.10.

* Use explicit version requirements for numpy

* Add documentation explaining test_zip_properties fix history

Document that this test was previously skipped for Python 3.13 due to
numpy private API usage in check_numpy(), and reference commit 50e206b
which fixed the issue by switching to public numpy.lib.format APIs.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

---------

Co-authored-by: Dan Guido <dan@trailofbits.com>
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>

Author: 3 people

pyproject.toml

@@ -31,9 +31,9 @@ requires-python = ">=3.10"
 torch = [
     "torch >= 2.1.0",
     "torchvision >= 0.24.1",
-    "numpy >= 1.24.0; python_version < '3.14'",
-    "numpy >= 2.3.5; python_version >= '3.14'",
-    ]
+    "numpy >= 2.2.6, < 2.3; python_version == '3.10'",
+    "numpy >= 2.3.5; python_version >= '3.11'",
+]
 lint = [
   "ruff >= 0.8.0",
   "mypy >= 1.10.0",

test/test_polyglot.py

@@ -309,7 +309,8 @@ def test_torchscript_properties(self):
         }
         self.assertEqual(properties, proper_result)
 
-    @unittest.skip("FIXME: Failing for python 3.13")
+    # Previously skipped for Python 3.13 due to numpy private API usage in check_numpy().
+    # Fixed in commit 50e206b by switching to public numpy.lib.format APIs.
     def test_zip_properties(self):
         properties = polyglot.find_file_properties(self.zip_filename)
         proper_result = {