[WIP] support for generic network configuration

assigned to @anayak

Currently, complexity of numerical differentiation (to compute the Jacobian of non-linear functional) is non-linear and takes the most computational time. It would be effective to include automatic derivatives.

Making a sub-task to implement automatic-differentiation (branch add-autodiff).

mentioned in merge request !3 (merged)

The structure for implementing automatic-differentiation is a little more clear with the current version of single_pipe demo and the following conclusions are apparent in this strategy.

The key objective is "templating" the operator() functions within SteadyCompressorSystem and TransientCompressorSystem classes. Then solver for the network needs to instantiate the state-independent operators ONCE and the state-dependent operators should resolve the template type during the first run.

Currently the network instance instantiates these operators statically and this needs some changing since the template types are only resolved at compile-time. It is proposed that the network object be separated from the operator initialization bits of the implementation.

The easiest strategy would be to have an intermediate object, say "Model" which is instantiated statically within the operator() function and this handles the operator instantiation parts. A branch add-model-class will be spawned for this sub-task.