Update codegen for delay

[Martin Sjölund]

I have a patch where I have done some of the work to make delay() work more as expected.

But I ran into problems because during initialization, we call functionDAE with initial() set to false. And before storing the first delay-values. I was under the impression that we had (finally) replaced all calls to delay() with the actual expression during initialization. And of course I don't know how to fix initialization, so... Lennart maybe you want to have a look?

There are some bad diffs in my code (like changing ASSERT to assert to debug easier).

Basically, at runtime I want to call delayImpl(data, index, curTime) for most calls to delay(exp, T) (most things are known).
One thing that remains to be done is limiting the solver step size to at most T (should be an easy fix).

[Martin Sjölund]

Postponed until 1.9.1

[Lennart Ochel]

I think the initialization handles delay correctly. If there are still issues, please reopen this ticket.