Throw helpful error when unknowns in the initialization system don't have guesses

[hersle]

Attempt to fix #2873.

First draft is very much a WIP. Want to see which tests fail.

Checklist

• Appropriate tests were added
• Any code changes were done in a way that does not break public API
• All documentation related to code changes were updated
• The new code follows the
  contributor guidelines, in particular the SciML Style Guide and
  COLPRAC.
• Any new documentation only uses public API

[ChrisRackauckas]

Thanks, I did notice in a few cases we were missing an error message somewhere but didn't track down the best spot to put it. This looks correct.

[hersle]

Yeah I'm also not 100% sure where to put it. I guess this applies more generally to any NonlinearSystem, so maybe it should go there?

[ChrisRackauckas]

But an error at that level might be confusing because people might not understand the lack of contextualization, i.e. it would say "u0 is missing a value", when it means "u0 of the initialization" which means "guess". So I think this has to be specialized.

[hersle]

Maybe. The stacktrace should be very indicative of the context in such a case, though.

[hersle]

I am doubtful that this is a good way to fix the issue, at all. It would be better to hit the old error again, instead of throwing a new error a new place. Also, get_u0() does some processing further down the pipe that extends the list of guesses from that in InitializationProblem(), so it is not really correct to throw an error at such a high level. See comment in #2873.