[SQL] Support RANK, ROW_NUMBER, DENSE_RANK even if not in a TopK pattern

I think these operators can be compiled into rolling aggregates, e.g., 

```
SELECT  rank() OVER (ORDER BY col),
```

can be 

```
count(*) over (ORDER BY col RANGE BETWEEN UNBOUNDED PRECEDING and CURRENT ROW)
```

`dense_rank` corresponds to `count(distinct)`.

`row_number` however is more tricky with non-unit weights, so that one may still need a custom implementation.