bug(exprs): chain mutate() for window op and non window op throws error for pyspark backend

[LeeTZ]

Currently, chain mutate() for window will fail for pyspark backend:

table = client.table('time_indexed_table')
    context = (
        pd.Timestamp('20170102 07:00:00', tz='UTC'),
        pd.Timestamp('20170105', tz='UTC'),
    )
    window = ibis.trailing_window(
        preceding=ibis.interval(hours=1), order_by='time', group_by='key'
    )
result_pd = (
        table.mutate(
            count_1h=table['value'].count().over(window),
        ).mutate(count=table['value'].count())
        .execute(timecontext=context)
    )

failed with AnalysisException:

E               pyspark.sql.utils.AnalysisException: "grouping expressions sequence is empty, and 'time_indexed_table.`time`' is not an aggregate function. Wrap '(count(time_indexed_table.`value`) AS `count_yoyo`)' in windowing function(s) or wrap 'time_indexed_table.`time`' in first() (or first_value) if you don't care which value you get.;;\nProject [time#19, key#20L, value#21, count#52L, count_yoyo#53L]\n+- Project [time#19, key#20L, value#21, count_yoyo#53L, _w0#54L, count#52L, count#52L]\n   +- Window [count(value#21) windowspecdefinition(key#20L, _w0#54L ASC NULLS FIRST, specifiedwindowframe(RangeFrame, currentrow$(), 3600)) AS count#52L], [key#20L], [_w0#54L ASC NULLS FIRST]\n      +- Aggregate [time#19, key#20L, value#21, count(value#21) AS count_yoyo#53L, cast(time#19 as bigint) AS _w0#54L]\n         +- Filter ((time#19 >= 1483340400000000) && (time#19 < 1483405200000000))\n            +- SubqueryAlias `time_indexed_table`\n               +- LogicalRDD [time#19, key#20L, value#21], false\n"

[cpcloud]

@LeeTZ Can you still reproduce this?