feat(postgres): Support DIV() func for integer division

[VaggelisD]

Fixes #3601

This PR introduces support for Postgres's DIV() function which performs integer division. It mirrors the BigQuery implementation which stores the args in exp.IntDiv and generates it back to the function representation.

Postgres Div

[georgesittas]

Both this and BigQuery's implementation are not correct besides roundtripping.

Specifically, the return type is not necessary an integer in either of these dialects (BQ docs, Postgres docs), so the semantics may change when transpiling DIV to other dialects.

For example, try transpiling this from Postgres into SQLite and you'll notice that the source and resulting queries produce different results:

-- Postgres, results in 0.5 because the type of the RHS is NUMERIC and so we're actually doing a float division
SELECT 1 / DIV(4, 2)

-- SQLite, results in 0
SELECT 1 / CAST(4 / 2 AS INTEGER)

[georgesittas]

Transpiling from BigQuery may work because it doesn't use typed division semantics, but that's coincidental.

[VaggelisD]

This is still a WIP, will attempt to implement the proper semantics for exp.IntDiv in BQ/Duckdb/Postgres as well such that transpiling between each other preserves the result type.

[VaggelisD]

Sorry for the force pushes, had to resolve conflicts and also amended some changes by accident. Leaving the following for future reference:

From my search, it is Postgres, DuckDB and BigQuery which support integer division. Out of those, Postgres's DIV() always results in a DECIMAL; In DuckDB and BigQuery, the types of the operands can alter the result type.

For this reason, it's not possible to safely transpile DuckDB's/BigQuery's exp.IntDiv without type inference. As for Postgres, this PR always casts exp.IntDiv to a decimal and removes the cast if it's roundtripped.

[georgesittas]

The CAST(4 AS REAL) here looks a bit weird, I don't think it should've been added. The root cause is that the IntDiv generator doesn't properly populate the typed and safe args, disregarding the dialect's semantics. It seems like that method was added a long time ago, so this is most likely like an oversight.

[georgesittas]

I don't think it should've been added

This is because Postgres uses typed division, so since we're generating a Div as part of IntDiv's generation, the typed division semantics should be respected. See for example how we don't add a cast to REAL here:

>>> import sqlglot
>>> sqlglot.transpile("1/2", "postgres", "sqlite")
['1 / 2']

[georgesittas]

May be best to address this in a followup PR after all. I played around a bit and it seems like we'll need to think more about how to parse / generate IntDiv.