Feature: stop time adaptivity

[drreynolds]

This PR adds optional routines to the SUNDIALS time integrators so that time steps which are shortened to adhere to "stop time" requests will be disregarded from the time step and method order adaptivity processes.

This feature is disabled by default; if after extensive testing we wish to enable this by default, I recommend that this wait until the next major release.

[gardner48]

Should we make this a threshold instead of a boolean? So the check would be something like the following

sunrealtype eta_stop = h_stop / h;
if (eta_stop < skip_adapt_stop_time_eta_threshold)
{ 
  skip_adapt_stop_time = SUNTRUE;
}
else 
{
  skip_adapt_stop_time = SUNFALSE;
}

so a threshold of 1 skips adaptivity if the stop time reduced the step size and 0 always adapts (the current default behavior).

[gardner48]

In a future release, I think a better default might be 1/eta_max, so any stop time step that can't return to the old step size is ignored.

[gardner48]

Inside IDAStep set tstoplimited = SUNFALSE if the step with stop time step size needs to be reattempted

[gardner48]

Same comments as in IDA

[drreynolds]

In updating this PR, I updated the X_test_tstop.c unit test for each package to examine this feature, on the simplest-possible test problem y'(t) = 1, y(0)=0. Strangely, while the option indeed reduced the number of time steps with IDA, with CVODE the number of time steps was unchanged, and with ARKODE it increased. While I think that this feature is important, it is clear to me that its implementation will require more work, so I'm rescinding the PR for now. After this release, I'll try to take time to fix the implementation, and correctly add support for the threshold-based approach.