Fix visit_label override to handle nested labels

[JacobHayes]

Fixes #39 - passes the "test case" I posted there and a number of internal queries I've tested with. Will look into adding a formal test in the next few days.

Print:
SELECT sum(sum(`test`.`val`)) OVER () AS `anon_1`, sum(CASE WHEN %(param_1)s THEN `test`.`val` END) AS `sum_1`, sum(sum(CASE WHEN %(param_2)s THEN `test`.`val` END)) OVER () AS `anon_2`
FROM `test`

Compile:
SELECT sum(sum(`test`.`val`)) OVER () AS `anon_1`, sum(CASE WHEN %(param_1)s THEN `test`.`val` END) AS `sum_1`, sum(sum(CASE WHEN %(param_2)s THEN `test`.`val` END)) OVER () AS `anon_2`
FROM `test`

Literal compile:
SELECT sum(sum(`test`.`val`)) OVER () AS `anon_1`, sum(CASE WHEN true THEN `test`.`val` END) AS `sum_1`, sum(sum(CASE WHEN true THEN `test`.`val` END)) OVER () AS `anon_2`
FROM `test`

Result:
((6, 6, 6),)

I haven't noticed any regressions with expressions in the group by. That seems to be fully supported by BQ anyway, so I'm guessing the method I removed was just for nicer looking "group by" statements? I'll try to break this a bit more, but would love some other insight there.

[mxmzdlv]

Heya! This breaks tests test_custom_expression and test_compiled_query_literal_binds.

For example, this generated query:

SELECT TIMESTAMP_TRUNC(timestamp, DAY) AS `timestamp_label`, sum(`test_pybigquery.sample`.`integer`) AS `sum_1`
FROM `test_pybigquery.sample`
GROUP BY TIMESTAMP_TRUNC(timestamp, DAY)
ORDER BY sum(`test_pybigquery.sample`.`integer`)

Fails with SELECT list expression references column timestamp which is neither grouped nor aggregated at [1:24]

If we group by the label timestamp then it works, but also if we remove ORDER BY clause, e.g.:

SELECT TIMESTAMP_TRUNC(timestamp, DAY) AS `timestamp_label`, sum(`test_pybigquery.sample`.`integer`) AS `sum_1`
FROM `test_pybigquery.sample`
GROUP BY TIMESTAMP_TRUNC(timestamp, DAY)

Then it works too. A bit confusing.

[JacobHayes]

Hey @mxmzdlv thanks for the patience on this one! Finally got around to digging in a little more and think I found a simple solution.

[tswast]

Thanks for your patience.

I've pushed a commit with the test cases from your issue and confirmed that they are fixed by this PR.