High-level Simulation type to manage time stepping

[ali-ramadhan]

This PR introduces a Simulation type that manages time stepping with a high level interface.

I think diagnostics and output_writers will have to become keyword arguments to time_step!.

It will be a pretty big breaking change once we move diagnostics and output_writers outside of Model so would be good to agree on the design and API before refactoring.

@glwagner Let me know what you think, we can iterate on the design in this PR.

Resolves #432
Resolves #447

[glwagner]

Ah so great. One quick thought: we can reinterpret time_step to refer only to a single time-step, and move the current multi step function to run! Then time_step does not take keyword arguments. Is there a convenience function that takes AbstractModel as input rather than Simulation? I think that’s important for simple scripts and examples. Agree this is a huge breaking change that we should finalize as soon as possible.

[ali-ramadhan]

From discussion with @glwagner:

1. Should get rid of
   https://github.com/climate-machine/Oceananigans.jl/blob/73bea229721624d65d4e2b2b79810622cf221993/src/TimeSteppers/adams_bashforth.jl#L39
2. Multiple stop criteria would be nice: called via stop_simulation(simulation).
3. stop_criteria should be a Tuple of functions.
4. We want a run_until!(model, t) or time_step_until!(model, t) function for AB2 (and RK3).
5. Should be progress(simulation).
6. Deprecate but keep around time_step! for multiple time steps.

[codecov]

Codecov Report

Merging #621 into master will increase coverage by 0.34%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #621      +/-   ##
==========================================
+ Coverage   74.53%   74.87%   +0.34%     
==========================================
  Files         117      118       +1     
  Lines        2246     2277      +31     
==========================================
+ Hits         1674     1705      +31     
  Misses        572      572              

Impacted Files	Coverage Δ
src/Diagnostics/nan_checker.jl	33.33% <0.00%> (-46.67%)	⬇️
src/Utils/pretty_time.jl	40.00% <0.00%> (-13.34%)	⬇️
src/Utils/time_step_wizard.jl	100.00% <0.00%> (ø)	⬆️
src/Simulations.jl	93.75% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update eb9b07f...397c15d. Read the comment docs.

[ali-ramadhan]

Simulation API is almost ready.

Comments on some of the points above:
3. I made stop criteria an array so you can push/append extra stop criteria as needed.
4. Out of scope for this PR.
6. Went all out and completely removed time_step! for multiple time steps. Only the single step version remains.

Major breaking changes:

1. Diagnostics and output writers are now part of Simulation, not Model.
2. time_step! takes a single time step. For multiple time steps, use a loop or a Simulation.

Still need to:

1. Write some simple Simulation units tests although it's already used by a lot of tests already.
2. Update examples.
3. Update model setup documentation.

[glwagner]

Too much to review all the test edits, but this is epic. Onwards!