Collect and print/save timings

[johnomotani]

Useful to monitor for performance regressions.

Timers are printed at the end of a run (this can be disabled by setting display_timing_info=false in the [output] section), and saved to the output file. They can be plotted using makie_post_processing, see the docs.

Also, the Profile package seemed to be interacting badly with MPI in one case I tried - the run was much slower with profiling on, with a lot of time spent in MPI calls - so 'standard' profiling isn't necessarily representative and this feature provides an alternative. Hopefully this is a temporary bug in some external package(s) which will go away at some point, because sampling profiles are still pretty useful!

Provides an option to switch on extra 'debug timers', see the docs.

[johnomotani]

One thing to be aware of with this PR - in order to save all the timers that get created to the output files, we have to gather a list of them, possibly update the list for the first couple of output steps, and MPI-communicate the list to all the MPI ranks that have to write to the output file. The code that does all that could be a little bit fragile if something happens that wasn't anticipated when it was written. Hopefully in getting all the tests to pass I've worked out those kinks, but if you get errors that come from write_timing_data() in file_io.jl or from format_global_timer() in timer_utils.jl, then this PR was almost certainly the cause. At worst I think those errors would cause a crash, or write incorrect timing data, there shouldn't be anything in this PR that could modify physics output of the code!