unsupported on new macOS)

in C++17

included in external headers

"glm", etc etc.

Seeing funny behaviour where for nearly-Gaussian maps MELODIC changes the
value of `mmthresh`, making the combined maps inconsistent.
  However, the overall aim is simply to make the null of the DR maps `≈N(0,1)`.
Instead of using the full mixture model, we simply rescale the maps based on
robust estimators of the mean and standard deviation (i.e. that should be
relatively insensitive to the heavy tails).

Create a binary mask, and then multiply (i.e. rather than treating positive and
negative separately).
  Note that we also create `dr_stage2_${s}_thresh` maps now so these are more
readily accessible to the user.

Makes it easier to debug by leaving full mixture model outputs.

No longer seems needed - though a little unsure as to use in the first place???

The call to imglob seems to be altering the order of ICs. This removes the
combination of glob/imglob and replaces it with an explicit loop over ICs.

Sprout from newnew 2018-08-07 13:55:34 UTC Matthew Webster <mwebster@fmrib.ox.ac.uk> 'rewritten ols to avoid injection'
Cherrypick from master 2013-05-09 12:06:52 UTC Christian Beckmann <c.beckmann@donders.ru.nl> 'ready for 5_0_4':
    test.cc
Cherrypick from master 2016-11-10 15:52:29 UTC Matthew Webster <mwebster@fmrib.ox.ac.uk> '3.15 for 510':
    melodic.h
Cherrypick from ciftiBranch 2015-04-20 11:46:26 UTC Matthew Webster <mwebster@fmrib.ox.ac.uk> 'Conditional iconv':
    Makefile
    melodic.cc
Cherrypick from fsl-5_branch 2018-07-03 15:49:20 UTC Matthew Webster <mwebster@fmrib.ox.ac.uk> 'clarified text':
    meldata.h
    meloptions.h