Include timing information in provenance

[benjeffery]

Tracking wallclock and cpu time used in inference is challenging in the distributed inference pipelines we're running, as they are often interrupted with steps redone and the tooling around the pipeline unable to report time actually used by tsinfer accurately.
It would be possible to record this information in the provenance field of the tree sequence, steps that are distributed can store their timings and combine them on finalise.