Add timing for all intermediate GC phases to diagnose the 50-80ms gap
between timed phases and total gc.collect() time.

New timing fields:
- scan_heap_ns: parallel scan_heap phase
- disable_deferred_ns: disable deferred refcounting loop
- find_weakrefs_ns: find_weakref_callbacks
- stw1_ns: StartTheWorld python#1
- objs_decref_ns: cleanup_worklist(objs_to_decref)
- weakref_callbacks_ns: call_weakref_callbacks
- finalize_ns: finalize_garbage
- stw2_ns: StopTheWorld python#2
- resurrection_ns: handle_resurrected_objects
- freelists_ns: _PyGC_ClearAllFreeLists
- clear_weakrefs_ns: clear_weakrefs
- stw3_ns: StartTheWorld python#3

All values exposed via gc.get_parallel_stats()['phase_timing'].