change to dynamic linking and add interface and threading layers

[vtavana]

resolves IntelPython/mkl_fft#165

Main branch

import numpy as np, mkl_umath
n = 2**31
data = np.ones(n)
mkl_umath.absolute(data)
# Intel oneMKL ERROR: Parameter 1 was incorrect on entry to vdAbs.
# array([0., 0., 0., ..., 0., 0., 0.], shape=(2147483648,))

This branch

import numpy as np, mkl_umath
n = 2**31
data = np.ones(n)
mkl_umath.absolute(data)
# array([1., 1., 1., ..., 1., 1., 1.], shape=(2147483648,))

[ndgrigorian]

In general I don't see a problem with this change, but changing a dependency is a pretty big tweak.

@ekomarova @AndresGuzman-Ballen is there any way Intel NumPy tests can be run with this branch to make sure it doesn't introduce any major failures?

[AndresGuzman-Ballen]

@ndgrigorian I suppose I could point mkl_umath's commit in numpy-2.2.5 branch to this branch, but I'd prefer doing that after I can get all subpackages to build and pass tests in our CI