Add automated test coverage with PyPy

[bkeryan]

The README.rst and trove classifiers claim support for PyPy, but there is no automated test coverage of this configuration.

In #625 I tried updating the GitHub workflows to run unit tests with PyPy, and ran into a number of problems with dependencies (mostly grpcio, grpcio-tools, and numpy).

AB#3233994

[bkeryan]

I think we should be able to get the tests running on PyPy in library mode, but not gRPC mode. We also ought to add some caveats about gRPC/PyPy compatibility to the readme.

NumPy

NumPy appears to distribute binary wheels for one version of PyPy. For example, https://pypi.org/project/numpy/2.3.2/#files includes binary wheels for pypy3.11.

I got the nitypes tests running with multiple versions of PyPy by specifying PyPy-specific version constraints: https://github.com/ni/nitypes-python/blob/main/pyproject.toml#L38

numpy = [
  { version = ">=1.22", python = ">=3.9,<3.12", markers = "implementation_name != 'pypy'" },
  { version = ">=1.26", python = ">=3.12,<3.13", markers = "implementation_name != 'pypy'" },
  { version = ">=2.1", python = "^3.13", markers = "implementation_name != 'pypy'" },
  { version = ">=2.1", python = ">=3.10,<3.11", markers = "implementation_name == 'pypy'" },
  { version = ">=2.3", python = "^3.11", markers = "implementation_name == 'pypy'" },
]

Note that specifying these constraints in the [tool.poetry.dependencies] section affects the distribution package's metadata, which may require consumers of the package to specify the same constraints. It's better to specify these constraints in a dev dependency group.

The PyPy FAQ has some interesting info about NumPy compatibility: https://doc.pypy.org/en/latest/faq.html#what-about-numpy-numpypy-micronumpy

Protocol Buffers

The protobuf package has multiple implementations: pure Python, legacy pyext, and upb.

protobuf on PyPy appears to use the pure Python implementation:

PS C:\dev\nitypes-python> poetry run python
Python 3.11.11 (0253c85bf5f8, Feb 26 2025, 10:43:25)
[PyPy 7.3.19 with MSC v.1941 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>>> import google.protobuf.internal.api_implementation
>>>> google.protobuf.internal.api_implementation.Type()
'python'
>>>> google.protobuf.__version__
'6.32.0'

gRPC

The grpcio package is written in Cython and does not distribute binary wheels for PyPy. There is a GitHub issue about supporting grpcio on PyPy: grpc/grpc#4221