Emit major slice counters in the runtime events

[kayceesrk]

This PR emits the counters updated as part of the major slice computation. The related discussion is here: #13321. olly has already been updated to emit these counters in the trace: tarides/runtime_events_tools#46.

I had originally planned to make this PR after #13320 was merged. However, it looks like it may take longer: #13320 (comment). The change itself can be easily modified to accommodate future changes in the slice computation.

[gasche]

This is fine overall. I think it would be nice to try to document a bit the new runtime_counters constructors, possibly just to say that they correspond to internal implementation details of the GC related to pacing computation.

[gasche]

The constructors above all have careful documentation. Shouldn't we try to document those as well, or at least make it explicit why they are not documented?

[kayceesrk]

Good point. Let me try to document these.

[tmcgilchrist]

That would generate useful documentation on the Runtime_events module.
Unfortunately some of the other constructors are undocumented.

[sadiqj]

I've added some docs for the constructors in kayceesrk#16

One problem with this PR (ocaml/ocaml) though is that major_collection_slice can get hammered pretty frequently as we try to do opportunistic major slices in both in the minor GC entry and exit barriers. It is called in the spin wait loop and we had to add a flag to prevent emitting tens of thousands of phase events: https://github.com/ocaml/ocaml/blob/trunk/runtime/major_gc.c#L1643

We probably don't want to emit events when that happens. (do we even want to be updating our work in that case?)

[kayceesrk]

(do we even want to be updating our work in that case?)

Good point. I see that slice_target and slice_budget are updated in the case of opportunistic slices. We'd have to do something about them.

I'd prefer to keep logic changes in a separate PR. For now, I've updated update_major_slice_work to take in log_events as an argument. The caml_gc_log and caml_gc_message remain unguarded as is the case with major_collection_slice.

[gasche]

I am happy with the current state. Approved. Thanks @sadiqj for the review, suggestion to use log_events, and documentation.

[kayceesrk]

Thanks @gasche.