rgw: add pg_ver to tombstone_cache

[cbodley]

a tombstone cache was added to remember the mtime of deleted objects for
use with the HTTP_IF_MODIFIED_SINCE header, but the comparison was still
failing because of a missing pg_ver. added pg_ver to the tombstone cache
so it can be passed with HTTP_DEST_PG_VER

[theanalyst]

I'll run this by and see if I can see the old behaviour again

[cbodley]

@yehudasa confirmed that we don't compare pg_ver unless they're from the same zone:

  bool operator<(const obj_time_weight& rhs) {
    ...
    if (zone_short_id != rhs.zone_short_id) {
      return (zone_short_id < rhs.zone_short_id);
    }
    return (pg_ver < rhs.pg_ver);
  }

[yehudasa]

@cbodley perfect

[yehudasa]

@theanalyst did you test this?

[theanalyst]

Yeah ran this through the same script, no errors so far

[mattbenjamin]

The main questions I have are longer-term:

1. what allows us to rely on a memory-only cache?
2. can the cache size limit be overrun? what happens in that case?