fix: handle consuming streams with no data

[plamut]

Fixes #27.

This PR fixes the issue with consuming streams with no data. If an empty stream is encountered, the to_dataframe() / to_arrow() method returns an empty DataFrame / arrow Table.

The schema of the empty result is preserved (on a best-effort basis) and is consistent regardless of the chosen session data format.

How to reproduce

Run a query and fetch its results in an AVRO/ARROW session with multiple requested streams. The query results should be large enough so that the backend indeed decides to create multiple streams.

Additionally, the session should have a very tight row_restriction filter applied so that only a few rows actually get streamed to the client. If "lucky", at least one of the streams will contain no data and will result in an error when reading from it.

Things to discuss

• Do we need to backport this fix to v1beta1 client? I presume not?

PR checklist

• Make sure to open an issue as a bug/issue before writing your code! That way we can discuss the change, evaluate designs, and agree on the general idea
• Ensure the tests and linter pass
• Code coverage does not decrease (if any source code was changed)
• Appropriate docs were updated (if necessary)

[shollyman]

I think this is fine for this change, but I wonder if we should consider consolidating all our various type/schema conversion code in the storage library and bigquery. Is there demand outside of our own usages (e.g. in other storage APIs) that we should consider moving this into a more central dependency?

[plamut]

Maybe in some libs working closely with the BigQuery API? cc: @tswast

[tswast]

We have similar mappings in pandas-gbq and maybe ibis.

https://github.com/pydata/pandas-gbq/blob/d251db03b159447331ac9ae63e13d295d75bad70/pandas_gbq/schema.py#L71

[plamut]

I'll try to have a look at it in the near future. It might be worth to wait for all the pending fixes and dependency version updates, though, to see how of that "lipstick" logic for types is still needed.