- in addition to changing the to the use the iterated Gram-Schmidt
   algorithm from the greedycpp code, this also changes the validation
   and enrichment method, making it more efficient
 - the code in LALapps's ppe_roq.c has be modified to reflect these
   changes

 - in addition to changing the to the use the iterated Gram-Schmidt
   algorithm from the greedycpp code, this also changes the validation
   and enrichment method, making it more efficient
 - the code in LALapps's ppe_roq.c has be modified to reflect these
   changes

LALInferenceGenerateROQ.[ch]: Added functions to enrich the whole training set rather than directly enriching the reduced basis - the prevents some numerical precision issues seen from doing the latter
 - refs #4123
Original: 0d28a3f04addd71060053fddcbbab692e727482c

  - rather than passing gsl_matrix and gsl_vector variables to functions we now pass
    LALsuite Arrays and Vectors
  - enriching a reduced basis given new training data can be performed
  - the <h|h> quadratic terms are now assumed to be computed using a reduced basis
    formed from a training set of the quadratic
  - refs #4123
Original: 8bc5cb67293844584837d0280c137ce38f6e5138

LALInferenceGenerateROQ: added functions to generate an orthonormal basis from a training set and an ROQ interpolant.
 - closes #1742
Original: 1cca4d780c0d92c3a38a459fb1596c21acb4f83b