Phillip Cloud

@jcrist I can't remember, are we not already doing this?

@jitingxu1 This PR has a lot of failures. Can you take a look so we can decide how to move forward?

This PR is stale and had a number of unaddressed issues.

@jitingxu1 Can you describe the approach you're thinking about a bit?

The difference in performance is pretty stark: ``` ---------------------------------------------------------------------------------- benchmark 'test_dot_columns[1000000]': 2 tests --------------------------------------------------------------------------------- Name (time in ms) Min Max Mean StdDev Median IQR Outliers OPS Rounds Iterations ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- test_dot_columns[1000000]...

Another approach is to replace the use of `.columns` in internals, (perhaps with a private `_columns` property), which is annoying but also not a breaking change. The downside there is...

Moving to draft to avoid merging before 9.4

The 3 cloud tests that were failing are now passing. ``` ibis/backends/snowflake/tests/test_client.py::test_insert PASSED [ 33%] ibis/backends/bigquery/tests/system/test_client.py::test_parted_column[date] PASSED [ 66%] ibis/backends/bigquery/tests/system/test_client.py::test_parted_column[timestamp] PASSED [100%] ========================================================================================= 3 passed in 39.24s ========================================================================================= ```

We really need to straighten out the pyarrow versus duckdb cloud read behavior. My vote is to use DuckDB's readers and work with them to smooth out any rough edges.

Whatever we do, we can't continue with fall back behavior.