suggest add numpy to dependance of codon-jit so that it can be installed automatically

[l1t1]

pip install codon-jit didn't install numpy, but import codon failed due to lack of numpy

pip install codon-jit 
...
Successfully installed astunparse-1.6.3 codon-jit-0.4.6 cython-3.2.4 packaging-26.2 six-1.17.0 wheel-0.47.0

python ../test-codonjit.py
Traceback (most recent call last):
  File "/par/tpy3145/../test-codonjit.py", line 1, in <module>
    import codon
  File "/par/tpy3145/codonenv/lib/python3.14/site-packages/codon/__init__.py", line 7, in <module>
    from .decorator import jit, convert, execute, JITError, JITWrapper, _jit_register_fn, _jit_callback_fn, _jit
  File "/par/tpy3145/codonenv/lib/python3.14/site-packages/codon/decorator.py", line 13, in <module>
    import numpy as np
ModuleNotFoundError: No module named 'numpy'

[BI71317]

MRE

import codon
from time import time


@codon.jit(debug=True)
def codon_numpy_sum(n: int) -> float:
    import numpy as np

    arr = np.arange(n, dtype=np.float64)
    return ((arr * 2.0) + 1.0).sum()


def main() -> None:
    n = 1_000_000

    t0 = time()
    ans = codon_numpy_sum(n)
    t1 = time()
    print(f"[codon numpy only] result={ans:.1f} time={t1 - t0:.6f}s")


if __name__ == "__main__":
    main()

Result

After uninstalling CPython NumPy:

$ python3 -m pip uninstall numpy
Found existing installation: numpy 2.4.4
Uninstalling numpy-2.4.4:
  Would remove:
    /home/swchoi/.venvs/codon311/bin/f2py
    /home/swchoi/.venvs/codon311/bin/numpy-config
    /home/swchoi/.venvs/codon311/lib/python3.11/site-packages/numpy-2.4.4.dist-info/*
    /home/swchoi/.venvs/codon311/lib/python3.11/site-packages/numpy.libs/libgfortran-040039e1-0352e75f.so.5.0.0
    /home/swchoi/.venvs/codon311/lib/python3.11/site-packages/numpy.libs/libquadmath-96973f99-934c22de.so.0.0.0
    /home/swchoi/.venvs/codon311/lib/python3.11/site-packages/numpy.libs/libscipy_openblas64_-32a4b2a6.so
    /home/swchoi/.venvs/codon311/lib/python3.11/site-packages/numpy/*
Proceed (Y/n)? Y
  Successfully uninstalled numpy-2.4.4

Running the MRE fails before the Codon-compiled function body is actually used:

$ python3 codon_numpy_only_demo.py 
Traceback (most recent call last):
  File "/home/swchoi/src/test_code/python_codon_dsl/codon_numpy_only_demo.py", line 7, in <module>
    import codon
  File "/home/swchoi/src/codon/install/python/codon/__init__.py", line 7, in <module>
    from .decorator import jit, convert, execute, JITError, JITWrapper, _jit_register_fn, _jit_callback_fn, _jit
  File "/home/swchoi/src/codon/install/python/codon/decorator.py", line 13, in <module>
    import numpy as np
ModuleNotFoundError: No module named 'numpy'
(codon311) swchoi@swc-WIN:~/src/test_code/python_codon_dsl$ ```
Running the MRE fails before the Codon-compiled function body is actually used:

Root Cause

This failure happens even when the user code only intends to use Codon's native NumPy support inside the JIT-compiled function, not CPython's NumPy module directly.

Looking at codon/install/python/codon/decorator.py, the decorator currently refers to CPython NumPy scalar types:

pod_conversions = {
    type(None): "pyobj",
    int: "int",
    float: "float",
    bool: "bool",
    str: "str",
    complex: "complex",
    slice: "slice",
    np.bool_: "bool",
    np.int8: "i8",
    np.uint8: "u8",
    np.int16: "i16",
    np.uint16: "u16",
    np.int32: "i32",
    np.uint32: "u32",
    np.int64: "int",
    np.uint64: "u64",
    np.float16: "float16",
    np.float32: "float32",
    np.float64: "float",
    np.complex64: "complex64",
    np.complex128: "complex",
}

Because of this, importing codon through the Python JIT path implicitly depends on the CPython NumPy package being installed,

even if Codon Native NumPy is only imported inside the Codon-compiled function body.

My understanding is that these entries are used to map Python / NumPy scalar types to Codon's statically typed representation.

Since Python's builtin numeric types do not always encode the exact fixed-width type information needed here, NumPy scalar types are being used as convenient concrete type markers.

A possible direction may be to avoid requiring CPython NumPy at module import time by using another internal/builtin representation for fixed-width scalar type markers.

Alternatively, if this dependency is intentional, it may be worth documenting that codon-jit currently requires CPython NumPy for NumPy scalar type conversion support, or just add module dependency,

even when the compiled function body uses Codon's native NumPy implementation.