add scipy to runtime dependency

[vtavana]

In #179, we added helper functions directly from SciPy for expanding the functionality of mkl_fft.interfaces.scipy_fft module. This addition means that scipy is now a runtime dependency. In this PR it is added to the project as a runtime dependency.

[Copilot]

Pull Request Overview

This PR adds SciPy as a runtime dependency to support the newly imported helper functions from SciPy in the mkl_fft.interfaces.scipy_fft module.

• Add scipy to the core dependencies in pyproject.toml
• Remove scipy from the optional test extras
• Include scipy in both conda recipe metadata files

Reviewed Changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File	Description
pyproject.toml	Added scipy to dependencies and removed from tests
conda-recipe/meta.yaml	Added scipy under requirements
conda-recipe-cf/meta.yaml	Added scipy under requirements

[vtavana]

@ekomarova needs this PR to be merged to ensure wheel packages function correctly.
And to have a green CI for this PR, we need to first merge #190

@ndgrigorian @antonwolfy When you have time, please review them.

[ndgrigorian]

My only concern is, does this mean that Intel NumPy can no longer be installed without SciPy, if mkl_fft is being used? Is this something we should do intentionally?

@ekomarova does infra team have concerns about this?

[vtavana]

My only concern is, does this mean that Intel NumPy can no longer be installed without SciPy, if mkl_fft is being used? Is this something we should do intentionally?

Alternatively, I can modify mkl_fft/interfaces/__init__.py to have

try:
    import scipy.fft
except ImportError:
    pass
else:
    from . import scipy_fft

With this change, Intel NumPy should work fine without a dependency on scipy.
For standalone mkl_fft, there is no need to install scipy as well except for running tests (as before) or if someone wants to use mkl_fft.interface for scipy (new requirement).

[ekomarova]

My only concern is, does this mean that Intel NumPy can no longer be installed without SciPy, if mkl_fft is being used? Is this something we should do intentionally?

@ekomarova does infra team have concerns about this?

Since numpy depends on fft, and fft depends on scipy, then yes. But we only rely on the dependencies of these sources. If this can be changed, and making a dependency on scipy is not necessary for fft, then this will solve the problem with numpy. Then what @vtavana suggests here #187 (comment) would be the alternative solution

[ndgrigorian]

@vtavana
I think that's sensible, to avoid introducing a strict dependency.

An additional alternative is that tests that use SciPy could be skipped without it in the environment, but that isn't strictly necessary I think.

[vtavana]

superseded by #195