sql: Add overloads to ARRAY_AGG

[abhimadan]

As referenced in #15939, ARRAY_AGG erroneously returns anyelement as its fixed return type. This follows the convention of existing array builtin functions to have overloads for all non-array types, which allows the fixed return type to be known for each overload.

Release note (bug fix): None

[cockroach-teamcity]

This change is 

[CLAassistant]

All committers have signed the CLA.

[justinj]

that was much simpler than I expected!

---

Review status: 0 of 2 files reviewed at latest revision, 1 unresolved discussion, some commit checks pending.

---

docs/generated/sql/aggregates.md, line 4 at r1 (raw file):

<thead><tr><th>Function &rarr; Returns</th><th>Description</th></tr></thead>
<tbody>
<tr><td><code>array_agg(arg1: <a href="bool.html">bool</a>) &rarr; <a href="bool.html">bool</a>[]</code></td><td><span class="funcdesc"><p>Aggregates the selected values into an array.</p>

😢 this is so unfortunate @knz do you know offhand if we have any way of supporting very basic generic functions (not full HM but something that would handle the array situation)?

---

Comments from Reviewable

[knz]

Review status: 0 of 2 files reviewed at latest revision, 1 unresolved discussion, some commit checks failed.

---

docs/generated/sql/aggregates.md, line 4 at r1 (raw file):

Previously, justinj (Justin Jaffray) wrote…

😢 this is so unfortunate @knz do you know offhand if we have any way of supporting very basic generic functions (not full HM but something that would handle the array situation)?

You mean to generate the documentation? Yes we could enhance the doc generator.

If you mean avoid instantiating the overload explicitly, yes I suppose we can bake something ad-hoc but I am not happy at the thought. If we are to invest in typing we should figure out something principled first.

---

Comments from Reviewable

[abhimadan]

So when I tried to fix the failing tests, I encountered an ORM compatibility issue with sequelize. It seems as though it expects to receive a string and convert it to an array of strings, but after my change, it receives an array of strings and fails when it tries to call strings functions on it. I'm investigating this more closely now, but as a result of this, I can't merge this PR right now.

[abhimadan]

I managed to work around the ORM compatibility issue by retaining most of the old ReturnTyper logic so that array_agg still retains aliased type information in the return type. @justinj PTAL!

[justinj]

LGTM, but maybe add some extra tests for the aliased types, like NAME to make sure they still work (and will continue to work).

[abhimadan]

A concern I have with this now is that it effectively creates undocumented overloads for array_agg for types that are aliased to types with defined overloads. It might not be a big issue since this function's behaviour is fairly straightforward, but I wanted to at least mention it.