[REDO] Parallel: Refactor PSCardTable::scavenge_contents_parallel

`PSCardTable::scavenge_contents_parallel` is ~170 LOC and the its structure/logic is rather hard to follow.

The original attempt (JDK-8280783) failed because it contained this assert:

```
assert(*dirty_r == clean_card || dirty_r == iter_limit_r, "inv");
```

This assert in the original patch is too strong, `dirty_r` can belong to next stripe (will be cleared/set as part of processing that stripe), but still iterated by current stripe.

Weaken it to:
```
assert(*dirty_r == clean_card || dirty_r >= clear_limit_r);
```

IOW, the dereference of of `dirty_r` can be a data-race if `dirty_r >= clear_limit_r`.