-iter qmr2 by default

[alkichigin]

According to the manual, "QMR2 is expected to be faster than QMR in finite-precision arithmetic [81] and requires slightly less memory".
Right now testing (from 1 to 30 eV)
adda -shape sphere -size 300 -grid 30 -beam electron 300 20 0 0 1 -prop 0 0 -1 -eps 4 -sym enf -scat_matr none -pol cm -no_vol_cor -iter qmr2 -lambda 41.32806614120375 -m 1.5 0.0
and qmr2 converges faster (converges at all) than qmr or especially bcgs2.
In other cases qmr2 might be preferable than qmr as well.

[myurkin]

It's a good idea. The only caveat is that it may (potentially) affect the automatic tests (tests/2exec) that are performed prior to release. So they should be run after the default iterative solver is changed.