Improve timing, add Trace Event Format output

[sim642]

Changes

1. Copy CIL's Stats into Goblint as Timing and clean it up. Although as-is CIL barely uses Stats, there's a large number of usages in some CIL plugins, so I decided not to remove it from CIL itself, but just not use it in Goblint. This is similar to what has been done with tracing at some point: CIL still contains Trace, but Goblint contains an improved copy as Tracing.
2. Rename time to wrap, so it's Timing.wrap instead of Stats.time now.
3. Fix inconsistent intermediate timing printouts. Previously it was possible that the sum of subtree times of a tree was greater than that tree itself if that tree was still in the middle of timing and hadn't exited cleanly. Now printing also adds all the in-progress times.
4. Add and improve measured data (all can be individually toggled):
   1. CPU time now includes CPU time of child processes (preprocessor executions). Previously preprocessing time couldn't be accounted for (even if it was wrapped in timing) because waiting for subprocesses exiting suspends Goblint and it doesn't do any work.
   2. Wall time is now measured as well. This rarely makes a difference (especially with preprocessing CPU time accounted for), but there is one case: when jobs is used to run multiple preprocessors in parallel, then wall time (which is what really matters for interactive usage) will be less than CPU time (which is more indicative of actual work done).
   3. Allocated memory is now measured as well. This is an idea I got from the landmarks library.
   4. Call counting existed in CIL already, but was disabled by default and we never used it, but I've recently found it to be quite valuable information. Otherwise if something takes 1s, then it might seem like something very slow is happening, but maybe it just gets executed 10000 times instead of the expected 100.
5. Rename printstats to dbg.timing.enabled.
6. Timing is generalized into a functor, which allows multiple simultaneous timing hierarchies. This allows to separately time how long we analyze what functions, for example.
7. Add Trace Event Format (TEF) output (activated using dbg.timing.tef) for both our normal timings and function analysis times (aka my magic flamegraphs). Documentation for viewing this is included. In TEF additional metadata can be included for timed sections, e.g. parsing timings have their filename attached, so it's easy to find out which files take the longest to parse.

TODO

• Update bench repo scripts with printstats rename.