Fix bug in H2 line cooling

See merge request !25

Priyanka reported a floating point exception abort in one of her runs.
It came from a straight-up division in CoolH2_GA08.

Enclose all divisions in that function in safe_div calls.

21sep19

See merge request !24

document increase in limit to Fe 17 levels

See merge request !23

We updated the LAMDA model of CN to their hyperfine
split version.  Many CN lines that were multiplets
are now individual lines.  We now find one of the
lines and obtain its intensity rather than the full
intensity of the multiplet.  We would need to create
a number of create line blends to get the multiplet
intensity.  Those are inconvenient since then all lines
have the same label “Blnd” and so are hard to identify
in the output.

This change puts us back to the previous state with the
non-resolved model and the full multiplet intensity being
obtained.  The LAMDA masterlist file has both
versions present with a comment on how to go to
the fully hyperfine split model.

We need to look into the literature to see what is
actually observed.  Do most telescopes measure
the multiplet individual members?  What do most
observational studies do?

we should have given the change date and the reason for the change.

Occurred in the defintion of save_img_matrix mathods in save.cpp, where
some methods used the same names in their API as some class variables.
Fixed.

Reintegration of branch 'imgmatr' has led to longer execution times of
the test suite.  That is because in Atom_levenN(), copies of the rate
matrix and creation vector are currently created whether the images are
created or not.

Create copies on demand instead.  Fixed.

FITS files created when negative populations are computed carried the
wrong species in the file name.  The species used is the one specified
in the 'save arrays levels' command, if one given.  Instead, the name
of the failing species should be used.  Fixed here.

FITS images of each of the matrix and vectors that appear in the linear
algebra equation solved are currently created as a debugging aid when
negative populations are computed.  However, neither iso_level() nor
atom_levelN() output the original matrices, but their contents after
manipulation by LAPACK.

Store the original matrix and creation vector, and also the computed
(negative) populations.

By the way, on the interface of the img_matrix methods replace the file
prefix with a bool indicating whether negative populations are stored.
A suffix is now added by a new method, set_basename().  Guarantees
consistency of FITS file names from both functions mentioned above when
negative pops are computed.

Discrepancies discovered in FLT_IS_DBL run.  Normal runs do not trigger
any warnings.

The FITS file contains three images, one each for A, b, and x of the
A.x = b linear system that Cloudy solves.  Hazy1 now explains that.

iso_level() expects to be given boolean parameters about whether to
print/save the rate matrix.  These can be accessed from within the
function from the isosequence and element given.  Remove them.

Levels for the 'print/save arrays levels' commands cannot be resolved
during parsing, but only after the atomic data have been loaded.
Until now, the resolution occurred in cloudy.cpp immediately after the
data loading.

This is now absorbed in the calls that do print/save the arrays to
simplify cloudy.cpp.

Creating output files through the 'print' family of commands feels a
little awkward.  It is preferable to generate FITS or PPM images with
a subcommand of 'save', as done here.

Note, however, that this is not a proper 'save' command, since it
functions as a debugging aid.  It is therefore treated a little
differently than its counterparts both in parse_save.cpp, and in
save_do.cpp.