Substitution/inlining

@mattwala I'm planning on using dagrt to back grudge. To that end, I've been thinking that I'd have the time stepper and the finite element method each generate a DAG, and then I would inline the FEM DAG into the time stepper DAG at each RHS call.

Some possible wrinkles in this include:

• Oww, giant DAG (but we'll need to be able to scale anyway)
• Naming of functions added to function registry as to be specific to each FEM discretization (in case multiple are involved in an RHS)
• Pretty redundant. But then--do we really want function calls in dagrt's language?

I'll probably plow ahead with this. If you've got any thoughts, I'd love to hear them tough.