improve dtmin and unstable detection

[oscardssmith]

Followup to https://github.com/SciML/SciMLBase.jl/pull/692. It turns out that OrdinaryDiffEq sets the minimum dt to eps(t) which is relatively sensible, but that means that we should be failing with an error if we don't accept a step and we're already at the minimum dt.

This definitely needs some tests, but I think those tests need to be in OrdinaryDiffEq so they will be added as a separate PR once this merges.

[codecov[bot]]

Codecov Report

Attention: Patch coverage is 32.00000% with 17 lines in your changes are missing coverage. Please review.

Project coverage is 29.31%. Comparing base (2df59ce) to head (761a8c7). Report is 4 commits behind head on master.

Files	Patch %	Lines
src/integrator_interface.jl	32.00%	17 Missing :warning:

Additional details and impacted files

@@             Coverage Diff             @@
##           master     #693       +/-   ##
===========================================
- Coverage   41.40%   29.31%   -12.10%     
===========================================
  Files          55       55               
  Lines        4557     4551        -6     
===========================================
- Hits         1887     1334      -553     
- Misses       2670     3217      +547     

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[oscardssmith]

Re-running CI (now that it's morning so codecov will hopefully be happier and now that the OrdinaryDiffEq test that I think was the only real failure is fixed).