Refinement of matrix inversion

See merge request giacomo.mulas/np_tmcode!67

first working implementation of iterative refinement to assess and improve numerical stability of numerical inversion of am

iterative refinement now checks for convergence only on first wavelength, afterwards it assumes a fixed number of refinement iterations

Prepare Inclusions

See merge request giacomo.mulas/np_tmcode!66

both in magma and lapack. Still to test, so far it just compiles
without errors.

update paths for Intel compilers

See merge request giacomo.mulas/np_tmcode!65

- add valgrind to the docker image used by gitlab-runner for the pipeline

See merge request giacomo.mulas/np_tmcode!64