Implement top_n() lazily

[lionel-]

Closes #2848

@javierluraschi sparklyr now creates this request with your reprex:

Selecting by  X
<SQL> SELECT `X`
FROM (SELECT `X`, rank() OVER (ORDER BY `X` DESC) AS `zzz2`
FROM `test_data`) `kmanvoxmpk`
WHERE (`zzz2` <= 10.0)

[javierluraschi]

Looks good!

BTW. I already pushed a patch to CRAN for sparklyr and top_n, so no need rush this into CRAN, I'll revert the sparklyr patch after dplyr updates in CRAN.