Fix automake to include headers for generalized solver

See merge request !7

Some header files are only included when MPI is active (e.g. in the
generalized solver), thus the normal distcheck does not fail if they
are missing in the corresponding automake variables and are not
distributed.

precision_typedefs.h is needed by the generalized methods, but only
included when built with MPI. Thus the normal distcheck does not catch
this.

The GPU initialization is actually quite constly, e.g. on Minsky it
takes roughly 0.7s. That is hurting performance for small matrices.
Thus a check has been added and now GPU should be initialized only the
first time.

and my_prow (col) values