Added errors for -ve rates, moved macro_pops call out of elements loop

[jhmatthews]

Some more robust errors and checks in macro_pops, and also a correction of a small error in saha.c which caused macro_pops to be called more times than necessary.

[jhmatthews]

Changes:

matom.c:

• Errors reported if negative rates or populations are obtained in macro_pops
• Added test to check that populations are solution to rate matrix equation in macro_pops

my_linalg.h:

• Added the function declaration gsl_blas_dgemv. Couldn't use gsl/gsl_blas.h as was getting conflicting variables from the other functions.

These changes are all error reporting / diagnostic. No functional changes yet.

There will need to be some changes in how we clean levels for population inversions- e.g. only clean if there is a radiative transition between the level or similar, and we also need to work out how to implement collisional transitions to link the singlet and triplet states. Merging this, as that stuff requires more testing.