Set-returning UDFs

**Is your feature request related to a problem or challenge? Please describe what you are trying to do.**
I'm looking at add an inverted index on my data and use it as the first step in more complicated data transformations. There's particular optimization that I think might be better done outside of the DataFusion engine.

**Describe the solution you'd like**
I'm imagining something like the following, where `inverted_index_search` returns an arbitrary number of rows.
```
SELECT inverted_index_search('some query');
```
```
+------+
| id |
+------+
| 1 |
| 2 |
+------+
```

I'd like to be able to write a udf (or udaf, or something new) that can return as many rows as it needs.

**Describe alternatives you've considered**
I looked at having my UDF return an array, but once I had the array I could not figure out how to unnest it. Issue https://github.com/apache/arrow-datafusion/issues/212 looks to cover that, but I think it might be better to just have the function return multiple rows directly.

**Additional context**
Apologies if I missed something that is already available for this.

The issue names no implementation files or tests. Start by reviewing DataFusion's existing UDF and UDAF behavior, then read issue #212 about unnesting arrays. Done means a function such as inverted_index_search can return an arbitrary number of rows directly and the SQL example produces the requested result.

Written by the indexing model from the issue text.