feff8.20x6 with Lanczos recursion

[Alex Ankudinov <alex@leonardo.phys.washington.edu>]

Dear developers,

The new version feff8.20x6 is now available from CVS on tiziano. This will
 likely be close to the release version of feff8.2, unless new bugs are
found. We are still working on the new FEFF8.2 document and new license
and copyright details.

Changes from feff8.20x5:

 1) bug fixes:
     a) dimensions of arrays for parallel runs (nph vs. nphx)
     b) format statement fix in scfdat.f for Mac compiler by Francois Farges
     c) new versions of padlib.f and padlib.h from Matt Newville
        to fix problems with padlib on CRAY machines.
     Thanks to J. Sims, Matt Newville, F. Farges, for help in troubleshooting
     these problems.  The new code now compiles and runs on all machines we've
     tested it on, including many parallel platforms.

 2) New fast recursion method FMS algorithm (BiCGStab) has been implemented.
    This is important for calculations of XANES for big clusters. 
    For example Si 600 atoms with lx=2 runs 5.5 faster with BiCGStab than
    with LU, and Cu 200 atoms, lx=3 runs 6 times faster than with LU.
    Typical (recommended) usage for BiCGStab.
    FMS  rfms2 [lfms2] [minv   toler1  toler2]
    FMS   14.0  0        1      0.001   0.001
    More details are given in the updated Feff8 doc. Further improvements
    in storage requirements are possible, but require many coding changes and 
    will be reserved for future developments.

 3) New extension to the MULTIPOLE card, in response to a suggestion
    from Michel Jaouen
     MULTIPOLE  le2 [l2lp]
      default: le2=0 (dipole only) and l2lp=0 (both l-->lp = l+/-1 channels)
      le2 = 2  adds quadrupole transitions (works only with l2lp=0)
      l2lp  = +1 include only dipole l-->l+1 transitions
      l2lp  = -1 include only dipole l-->l-1 transitions

  Please let us know if you spot any problems with this version.

Alexei Ankudinov
J. Rehr