Mkl-fft

=====

Includes bug fix 54

=====

Due to removal of deprecated real-to-real FFT with `DFTI_CONJUGATE_EVEN_STORAGE=DFTI_COMPLEX_REAL` and `DFTI_PACKED_FORMAT=DFTI_PACK`
from Intel(R) Math Kernel Library, reimplemented `mkl_fft.rfft` and `mkl_fft.irfft` to use real-to-complex functionality with subsequent
copying to rearange the transform as expected of `mkl_fft.rfft`, with the associated performance penalty. The use of the real-to-complex
transform improves multi-core utilization which may offset the performance loss incurred due to copying.

=====

Added `scipy.fft` backend, see 42. Fixed 46.


Python 3.7.5 (default, Nov 23 2019, 04:02:01)
Type 'copyright', 'credits' or 'license' for more information
IPython 7.11.1 -- An enhanced Interactive Python. Type '?' for help.

In [1]: import numpy as np, mkl_fft, mkl_fft._scipy_fft_backend as mkl_be, scipy, scipy.fft, mkl

In [2]: mkl.verbose(1)
Out[2]: True

In [3]: x = np.random.randn(8*7).reshape((7, 8))
...: with scipy.fft.set_backend(mkl_be, only=True):
...: ff = scipy.fft.fft2(x, workers=4)
...: ff2 = scipy.fft.fft2(x)
MKL_VERBOSE Intel(R) MKL 2020.0 Product build 20191102 for Intel(R) 64 architecture Intel(R) Advanced Vector Extensions 2 (Intel(R) AVX2) enabled processors, Lnx 2.40GHz intel_thread
MKL_VERBOSE FFT(drfo7:8:8x8:1:1,bScale:0.0178571,tLim:1,desc:0x5629ad31b800) 24.85ms CNR:OFF Dyn:1 FastMM:1 TID:0 NThr:16,FFT:4

In [4]: np.allclose(ff, ff2)
Out[4]: True

======

Changed tests to not compare against numpy fft, as this broke due to renaming of `np.fft.pocketfft` to
`np.fft._pocketfft`. Instead compare against naive realization of 1D FFT as a sum.

Setup script is now aware of `MKLROOT` environment variable. If unset, NumPy's mkl_info will be queried.

======

Fixed unreferenced bug in `irfftn_numpy`, and adjusted NumPy interfaces to change to pocketfft in NumPy 1.17

======

Issue 39 fixed (memory leak with complex FFT on real arrays)