Draft: QMCkl

[addman2]

I am not a fan of a big changes but this is a big one. Many things are intertwined and was hard to deliver them separately.

1. Makefun.f90 was (partly) refactored. Now separate orbital are stored indevel_tools/makefun_factory. executing construct.py one can build makefun.
2. Each refactored orbital has now a header that looks like this:

  1   ! s orbital
  2   !
  3   ! - angmom = 0
  4   ! - type = Gaussian
  5   ! - normalized = yes
  6   ! - angtype = spherical
  7   ! - npar = 1
  8   ! - multiplicity = 1
  9   !
 10   ! = N * R
 11   !
 12   ! where N is the normalization constant
 13   ! N = (2*alpha/pi)**(3/4)
 14   !
 15   ! and R is the radial part
 16   ! R = exp(-alpha*r**2)
 17   !

Consisting of title line 1. Parameters out of which only angmom, npar and multiciplity are necessary (lines 3 -8). Continued by any doc string (lines 10 onwards). These headers than can be used for generating documentation.
3. New test added for makefun testing:

• index updates
• calculating solo values
• calculating values for pseudo averaging
• calculating gradients and laplacian

Here is a sample output:

4: ######################################
4: Checking orbital index  97
4:
4: Test: index movement
4: indorb   =  36 OK
4: indshell =  36 OK
4: indpar   =   1 OK
4:
4: Test: single value
4: tests/failed =   3/  0
4:
4: Test: single value gradient and laplacian
4: tests/failed =   4/  0
4:
4: Test: calculate values for pseudo average
4: tests/failed =   1/  0
4:
4:
4: Result = T
4:
4: ######################################
4: Checking orbital index  98
4:
4: Test: index movement
4: indorb   =  45 OK
4: indshell =  45 OK
4: indpar   =   1 OK
4:
4: Test: single value
4: tests/failed =   3/  0
4:
4: Test: single value gradient and laplacian
4: tests/failed =   4/  0
4:
4: Test: calculate values for pseudo average
4: tests/failed =   1/  0
4:
4:
4: Result = T
4:
4: ######################################

5. Added implementation of Cartesian orbitals with indexes 90 - 99, representing s, p, d, ..., m. This is important for QMCkl, because this library cannot do AD so there is no backwards algorithm. So I added cartesians to makefun and generated new makefun_b (with tapenade 3.3). The idea is to then QMCKkl in every place where is placable speeding up the calculation. Turbo is compilable with both QMCkl and QMCkl_gpu, but I am changing the implementation so QMCkl it self will not work now. Hopefully today it will be.

[kousuke-nakano]

Dear @addman2, hi! I wonder if I could merge this pull-request. Is it still in progress?