Error: ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject

## How to reproduce the behaviour
import en_core_sci_sm
File "/Users/melaniev/Documents/code/PDBe_public_repos/IUCr-annotation-pipeline/.venv/lib/python3.11/site-packages/en_core_sci_sm/__init__.py", line 2, in
from spacy.util import load_model_from_init_py, get_model_meta
File "/Users/melaniev/Documents/code/PDBe_public_repos/IUCr-annotation-pipeline/.venv/lib/python3.11/site-packages/spacy/__init__.py", line 6, in
from .errors import setup_default_warnings
File "/Users/melaniev/Documents/code/PDBe_public_repos/IUCr-annotation-pipeline/.venv/lib/python3.11/site-packages/spacy/errors.py", line 3, in
from .compat import Literal
File "/Users/melaniev/Documents/code/PDBe_public_repos/IUCr-annotation-pipeline/.venv/lib/python3.11/site-packages/spacy/compat.py", line 39, in
from thinc.api import Optimizer # noqa: F401
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/melaniev/Documents/code/PDBe_public_repos/IUCr-annotation-pipeline/.venv/lib/python3.11/site-packages/thinc/api.py", line 1, in
from .backends import (
File "/Users/melaniev/Documents/code/PDBe_public_repos/IUCr-annotation-pipeline/.venv/lib/python3.11/site-packages/thinc/backends/__init__.py", line 17, in
from .cupy_ops import CupyOps
File "/Users/melaniev/Documents/code/PDBe_public_repos/IUCr-annotation-pipeline/.venv/lib/python3.11/site-packages/thinc/backends/cupy_ops.py", line 16, in
from .numpy_ops import NumpyOps
File "thinc/backends/numpy_ops.pyx", line 1, in init thinc.backends.numpy_ops
ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject

Same error also happens with:

python -m spacy info --markdown

Traceback (most recent call last):
File "", line 189, in _run_module_as_main
File "", line 148, in _get_module_details
File "", line 112, in _get_module_details
File "/Users/melaniev/Documents/code/PDBe_public_repos/IUCr-annotation-pipeline/.venv/lib/python3.11/site-packages/spacy/__init__.py", line 6, in
from .errors import setup_default_warnings
File "/Users/melaniev/Documents/code/PDBe_public_repos/IUCr-annotation-pipeline/.venv/lib/python3.11/site-packages/spacy/errors.py", line 3, in
from .compat import Literal
File "/Users/melaniev/Documents/code/PDBe_public_repos/IUCr-annotation-pipeline/.venv/lib/python3.11/site-packages/spacy/compat.py", line 39, in
from thinc.api import Optimizer # noqa: F401
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/melaniev/Documents/code/PDBe_public_repos/IUCr-annotation-pipeline/.venv/lib/python3.11/site-packages/thinc/api.py", line 1, in
from .backends import (
File "/Users/melaniev/Documents/code/PDBe_public_repos/IUCr-annotation-pipeline/.venv/lib/python3.11/site-packages/thinc/backends/__init__.py", line 17, in
from .cupy_ops import CupyOps
File "/Users/melaniev/Documents/code/PDBe_public_repos/IUCr-annotation-pipeline/.venv/lib/python3.11/site-packages/thinc/backends/cupy_ops.py", line 16, in
from .numpy_ops import NumpyOps
File "thinc/backends/numpy_ops.pyx", line 1, in init thinc.backends.numpy_ops
ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject

## Your Environment

* Operating System: MacOS (26.5.1 (25F80)
* Python Version Used: 3.11.14
* spaCy Version Used: 3.6.1 or 3.7.5
* Environment Information: virtual env created with uv

Previous discussions:
https://github.com/explosion/thinc/issues/939
https://github.com/explosion/spaCy/issues/13875

I get the same error as described in the github issues above. For me this happens when trying to import either of the two English dictionary versions below.
en-core-sci-sm==0.5.3 (from https://s3-us-west-2.amazonaws.com/ai2-s2-scispacy/releases/v0.5.3/en_core_sci_sm-0.5.3.tar.gz)
en-core-sci-sm==0.5.4 (from https://s3-us-west-2.amazonaws.com/ai2-s2-scispacy/releases/v0.5.4/en_core_sci_sm-0.5.4.tar.gz)
with thinc versions thinc==8.1.12 or thinc==8.2.4 and numpy==2.4.6

Reproduce the import of en_core_sci_sm and `python -m spacy info --markdown` in the reported Python 3.11.14 environment with spaCy, thinc, and numpy versions listed in the issue. Start by comparing the compatibility reports in thinc#939 and spaCy#13875; done means the model imports and the spaCy information command runs without the binary incompatibility error.

Written by the indexing model from the issue text.