ENH: signal: add CuPy/JAX delegation to scipy.signal

[ev-br]

Model the structure on scipy.ndimage:

• add _delegators.py with *_signature functions
• add _signal_api.py to collect "bare" imports from _private modules
• add _support_alternative_backends.py to decorate "bare" functions
• in __init__.py, import decorated names from _support_alternative_backends.py

This is still WIP for: no longer WIP

• only decorate functions, exclude lti classes, BadCoefficients etc
• double-check the namespace content for missing / extra items
• double-check, exclude the .windows subpackage
• make sure python dev.py -b all works without errors

[ev-br]

This is ready from my side, now that the delegation machinery skips classes, mypy noise is suppressed, and dev.py -s signal -b all seems to be working.

This PR, strictly speaking, does not bring anything new. It only streamlines the foundation for further array api work in scipy.signal. So it'd be most convenient to land this before other ones (gh-22122, gh-22008, gh-21783 will need to be rebased after this one lands).

EDIT: this PR actually does bring a blanket delegation to CuPy for all of scipy.signal namespace:

In [7]: import cupy as cp

In [8]: from scipy.signal import upfirdn
   ...: upfirdn(cp.asarray([1, 1, 1]), cp.asarray([1, 1, 1])).__class__
Out[8]: cupy.ndarray

In [9]: from scipy.signal import upfirdn
   ...: upfirdn([1, 1, 1], [1, 1, 1]).__class__
Out[9]: numpy.ndarray

the .windows namespace is separate (gh-21783), support for other backends is incomplete and will be taken care of in follow-up PRs. Support for lti/dlti and other classes will also be worked on in a follow-up. But all functions from scipy.signal delegate to their cupyx.scipy.signal analogs.

[lucascolley]

Looks great, thanks! I'll run GPU tests before merging

[lucascolley]

pixi run test-cuda -s signal is all green, thanks Evgeni!