re-expose refinenement precision

[kostrzewa]

Add back support for writing something like

BeginMonomial NDCLOVERRAT
  Timescale = 1
  kappa = 0.1400645
  CSW = 1.74
  AcceptancePrecision =  1e-21
  ForcePrecision = 1e-16
  StildeMin = 0.0000376
  StildeMax = 4.7
  MaxSolverIterations = 500
  Name = ndcloverrat_0_3
  DegreeOfRational = 10
  Cmin = 0
  Cmax = 3
  ComputeEVFreq = 0
  2Kappamubar = 0.0394421632
  2Kappaepsbar = 0.0426076209
  AddTrLog = yes
  UseExternalInverter = quda
  UseSloppyPrecision = single
  RefinementPrecision = half    # <- refinement precision                                                                                                                                                                                                                                                
  solver = cgmmsnd
EndMonomial

which allows the multi-shift solver to run up to single precision and then refines any unconverged shifts using double-half mixed CG, leading to another 20-25% improvement in time to solution for those monomials.

[kostrzewa]

@simone-romiti can you give this a test run to see if I didn't miss anything?

[kostrzewa]

@simone-romiti can you explain which parts of this you tested before merging? Did you try different combinations of sloppy and reconstruct precisions?

[simone-romiti]

I tried the 9 combinations with A, B = double,single,half in the input file:

... UseSloppyPrecision = A RefinementPrecision = B # <- refinement precision ...
#

[kostrzewa]

Perfect, thank you!