Don't redefine cblas/lapack routines from MKL

[d-torrance]

Also define MKL_Complex16 as double before including ffpack.h. If it was built with MKL support, then compilation will fail with a bunch of cannot convert 'double*' to 'MKL_Complex16*' errors.

After these changes, I was able to build M2 and FFLAS-FFPACK using MKL as their BLAS/LAPACK implementation.

Here are the steps to reproduce the issue. In particular, we need FFLAS-FFPACK to pull in the MKL headers. This is on Ubuntu 24.04 with intel-mkl installed:

sudo update-alternatives --config libblas.so-x86_64-linux-gnu
# select /usr/lib/x86_64-linux-gnu/libmkl_rt.so
cd M2/M2/BUILD/build
../../autogen.sh
../../configure CPPFLAGS="$(pkg-config --cflags blas)"
# need fflas-ffpack built w/ MKL support, and it only detects it using CBLAS_LIBS
make -C libraries/fflas_ffpack CBLAS_LIBS="$(pkg-config --libs blas)"
make

Closes: #3372

🤖 AI Disclosure 🤖

I did a bit of brainstorming with Claude, but the code is all mine.

[andrew-tawfeek]

This looks good. I compiled the affected FFLAS and mutable-matrix translation units both normally and with FFLAS forced through the MKL path against Intel MKL headers; the header guards suppress the conflicting declarations and the complex type matches the existing double-based storage.

AI-assisted PR review alongside GPT 5.6-Sol