Make sure not to work with possibly uninitialized variables

[kronbichler]

Due to a change in #14251, we would now not necessarily initialize a vector in the GMRES solver, see here:

dealii/include/deal.II/lac/solver_gmres.h

Line 598 in f28d4d1

data[i]->reinit(temp, true);

In two of our tests, we would use vector *= 0.0 to set its entries to zero, rather than the (more efficient) vector = 0.0. The former accesses whatever content we have in the memory, which could be NaN, which can then cause a floating point exception as in this test here: https://cdash.dealii.43-1.org/test/32205805 - while the latter is correct and makes sure we overwrite all content, including inf and nan data, with the correct value 0.