Add tspan option to DynamicSS

[tkf]

You can limit the solver by maxiters but sometimes it's handy to stop it via the simulated time tspan.

[tkf]

I feel dumb to do this. Is = safer? (e.g., when u0 happened to be a steady state) Or we need a callback that sets a flag? Maybe it's better to have a :Terminated return code which is set by terminate!.

[ChrisRackauckas]

== is safer. The handling should make this exact.

[tkf]

If we go with :Terminated return code #8 (comment), I guess we can do something like the following.

In DiffEqBase, define:

function __terminate! end

function terminate!(integrator)
  __terminate!(integrator)
  integrator.sol = solution_new_retcode(integrator.sol, :Terminated)
end

In OrdinaryDiffEq, define __terminate! instead of terminate! (and similar in Sundials):

function DiffEqBase.__terminate!(integrator::ODEIntegrator)
  integrator.opts.tstops.valtree = typeof(integrator.opts.tstops.valtree)()
end

Also, maybe define a simple function so that it's easy to support compatibility for common cases:

iserror(sol) = !(sol.retcode in (:Success, :Terminated))

[ChrisRackauckas]

Honestly I think the return codes should be enums, so if we redo them we should go all the way there.

[tkf]

Yeees! enum retcode would be really nice.

[ChrisRackauckas]

I'll merge when tests pass. But the retcode changes of adding a terminate and changing to enums is something that should get followup when we have time.

[ChrisRackauckas]

This should get docs