Add SuperLUMT support

[dweindl]

Closes #680.

Only for CMake-based build yet.

TODO:

• Add wrapper
• Build script for SuperLUMT, SuperLUMT option for sundials, CMake option in AMICI
• Test
• Document
• Allow setting number of threads
• NewtonSolverSparse::NewtonSolverSparse still uses SUNKLU
• Enable for Python?
• Enable on travis?
• Enable for Matlab?

[codecov]

Codecov Report

Merging #681 into develop will increase coverage by 0.01%.
The diff coverage is 63.63%.

@@             Coverage Diff             @@
##           develop     #681      +/-   ##
===========================================
+ Coverage    72.95%   72.97%   +0.01%     
===========================================
  Files           49       49              
  Lines         7041     7045       +4     
===========================================
+ Hits          5137     5141       +4     
  Misses        1904     1904

Flag	Coverage Δ
#cpp	69.58% <63.63%> (ø)	⬆️
#python	83.37% <ø> (+0.05%)	⬆️

Impacted Files	Coverage Δ
include/amici/solver.h	100% <ø> (ø)	⬆️
include/amici/newton_solver.h	100% <ø> (ø)	⬆️
src/steadystateproblem.cpp	82.22% <ø> (ø)	⬆️
include/amici/sundials_linsol_wrapper.h	45.45% <ø> (ø)	⬆️
src/sundials_linsol_wrapper.cpp	18.06% <ø> (ø)	⬆️
src/hdf5.cpp	88.02% <100%> (+0.02%)	⬆️
src/solver.cpp	70.64% <50%> (+0.11%)	⬆️
src/newton_solver.cpp	89.04% <85.71%> (-0.62%)	⬇️
python/amici/__init__.py	83.63% <0%> (+1.81%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 356fdde...bf4cb1f. Read the comment docs.

[dweindl]

First quick test with CS_Signalling_ERBB_RAS_AKT (1228 states) not too promising. Fastest with minDegATPlusA ordering. Using one thread, Simulation took 2x longer than with KLU. Using 2 threads doubles runtime. Should test with something bigger.

[dweindl]

I'd leave it as is for now. Can be used via C++ interface. The other points can be addressed once it's really needed. For me SuperLT_MT does not seem practical.