Floating point precision parameterization

[andrewreisner]

While the floating point precision is a paramter in the C++ control layer, local variables in the Fortran kernels have not been modified to use this parameter.

[andrewreisner]

The MSG library uses a preprocessor definition to set floating point precision. The floating point precision will need to be a build option.

[andrewreisner]

This parameterization will also need to be carried through any MPI calls outside the MSG library. For example:

1. line relaxation
2. redistributed coarse-grid solver
3. Cholesky and BoxMG coarse grid solvers