remove BGJ spaghetti code

[jeffhammond]

BGJ added a bunch of spaghetti that serves no purpose. This is the easiest and most important to remove.

[edoapra]

I am not sure this change is going to work since it is the one that initializes bgj_rtdb (QA tests might work with this PR if rtdb=0)
It might need to replaced by a call in nwchem.F to bgj_init(rtdb)

       subroutine bgj_init(rtdb)
       implicit none
#include "bgj_common.fh"
      integer rtdb
      bgj_rtdb = rtdb
      return
      end

[jeffhammond]

Yeah, it's definitely not working yet. I am in the process of ripping out a lot more stuff, and adding rtdb as an argument to many routines, to eliminate the BGJ spaghetti.

[jeffhammond]

@edoapra I ran the test suite and the results seem favorable but I know we don't have perfect test coverage. I was unable to get IBM XLF to -qextchk this the way I remember it doing in the past. Anyways, it might be worth inspection now.

Unfortunately, because of the way ga_iter_lsolve involves Fortran functions, I have to pass the rtdb argument via evil methods, but it is localized and theoretically can be fixed if we change everything (ROHF, UHF, ROKS, UKS, etc) the same way.

[jeffhammond]

@edoapra i think this is ready for review now. i ran the full test suite on it a while ago and it was fine.

[edoapra]

Some of the properties tests are still failing.
One reason could be that you have not changed the interface to some of the shell_fock_build variants (e.g. shell_fock_build_cam)

[jeffhammond]

ok i will look again. thanks. the test suite was running slowly.

[edoapra]

ok i will look again. thanks. the test suite was running slowly.

A couple of tests are pretty slow right now because they do convert 64_to_32 (against the full list of lapack routine) the whole source with the idea of detecting lapack calls.