Multiphase tracer produces singular matrix if saturation is zero

A case that is like in two-phase models is that one phase disappears. If the tracer only exists in this phase, the tracer also disappear, however that means that the local equation/residual degenerates.

This is automatically the case for all dofs where saturation is zero (over the whole time integration interval, see #703). A robust solution could be to check for zero saturation and replace the equation for those dofs with the Dirichlet constraint that the concentration is zero. For that it would need to be possible to set Dirichlet for every dof even when not on the boundary, see #704 (closed).