Fix single_distinct_to_groupby for arbitrary expressions

[james727]

Which issue does this PR close?

This addresses the bug in single_distinct_to_groupby.rs described in #1512

Right now the SingleDistinctToGroupBy optimizer rule only works on literal or column expressions. For more complex expressions (like binary operators), the optimization breaks due to searching for the right/left operands in the derived schema - where the relevant fields do not exist. This adds an alias to the group by expression and uses the alias to convert the outer expression to a column expression.

As an example - consider the query SELECT fn1(DISTINCT 2 * col), fn2(DISTINCT 2 * col) FROM t. Previously this was being rewritten as something like:

SELECT fn1(2 * col), fn2(2 * col)
  FROM (
    SELECT 2 * col FROM t GROUP BY 1
  )

This breaks, since converting the outer binary expression into a physical plan requires the subquery schema to have col as a column. This change rewrites the plan with an alias, as follows:

SELECT fn1(alias1), fn2(alias1)
  FROM (
    SELECT 2 * col AS alias1 FROM t GROUP BY 1
  )

What changes are included in this PR?

This just includes the logic fix and some tests that would have broken before.

Alternatives considered

We could also just disable this optimization for non literal/column expressions.

Input needed

Using a constant for the alias feels a little questionable but I think there's no risk of collisions based on how is_single_distinct_agg works. Open to suggestions though.

[alamb]

Thanks @james727 -- I'll take a look at this tomorrow!

[alamb]

I think this looks great -- thank you so much @james727 -- great first contribution

[alamb]

Thanks again @james727

[james727]

No problem, glad it's useful. Appreciate the review.