OPT_PLAN.txt

* monte_carlo.f90: added input block OPT_PLAN with run params for each opt cycle
  -- Pablo Lopez Rios, 2017-07-21

  OPT_PLAN allows specifying different parameters for each optimization cycle

  The block has one line per optimization cycle, each containing the cycle index
  followed by any number of '<keyword>=<value>' assignments, where current valid
  keywords are:
  * method: sets OPT_METHOD
  * jastrow: sets OPT_JASTROW
  * backflow: sets OPT_BACKFLOW
  * det_coeff: sets OPT_DET_COEFF
  * orbitals: sets OPT_ORBITALS
  * geminal: sets OPT_GEMINAL
  * maxiter: sets OPT_MAXITER
  * fix_cutoffs: determines whether to fix cut-offs or not, extending
    the functionality of OPT_NOCTF_CYCLES

  OPT_PLAN keywords not specified for any given cycle remain at the value set by
  the corresponding input keyword (or its default value if absent).

  The following example will run one cycle of varmin_linjas on the Jastrow
  parameters (excluding cutoffs by design), followed by one cycle of madmin on
  all Jastrow parameters, followed by two cycles of emin on all Jastrow
  parameters excluding cutoffs, followed by two cycles of emin on the Jastrow
  and backflow parameters excluding cutoffs:

   opt_method  : madmin
   use_jastrow : T
   use_backflow: T
   opt_jastrow : F
   opt_backflow: F

   %block opt_plan
   1 method=varmin_linjas
   2
   3 method=emin fix_cutoffs=T
   4 method=emin fix_cutoffs=T
   5 method=emin backflow=T fix_cutoffs=T
   6 method=emin backflow=T fix_cutoffs=T
   %endblock opt_plan

  The internal non_empty_{jastrow,bacfklow} flags have been tidied up, and the
  code should know to keep the Jastrow and backflow off until needed.  E.g., in
  the above example the first 5 VMC and 4 optimization cycles will have backflow
  off (provided the backflow block in the initial correlation.data is "empty").
  Note that the somewhat recent pjastrow initialization mode where alpha_1 is
  set to zero no longer sets the "non-empty" Jastrow factor flag; as a
  consequence autotest example heg_3d_fluid/complex_varmin now gives different
  results.

  This commit also changes the OPT_METHOD-dependent headings in the
  "out" file to simply say "optimization", modifies envmc to parse this 
  correctly, and updates runqmc to ignore OPT_* keyword inconsistencies in the
  presence of OPT_PLAN.