feat(hogql): groups support - part 1

[mariusandra]

Problem

HogQL is missing support for groups.

Changes

This is part 1 of such support. I decided to split it up into two PRs to get the easy wins out quickly, and not get stalled with support work.

• Refactors the huge database.py into smaller files. The new changes are in groups.py and cohort_people.py
• Adds fields goe_{index} to events to access "groups on events" properties. No support for "groups not on events" yet, see the TODO below.
• Splits raw_groups and groups tables. The table without "raw_" contains some SQL magic to remove duplicate fields.
• Do the same split for the "cohort_people" table because why not: it's been pending and was a small thing.

Todo in part 2

• Create an automatic join on the events table to access "groups not on events"
• Use short formats like "event.organization.bla" or "person.account.foo"
• Double check conflicts and team-id overrides with these short formats
• Make it work in non-HogQL insights

How did you test this code?

• I tested selecting from both the groups and cohort_people tables in the UI, with some changed rows in the database. Everything worked.
• Updated snapshots and queries in tests to show minimal changes.

[mariusandra]

Fixed the tests and this is ready for a review. Since the columns change for different tables with PoE on or off (person_id is added as a string field with PoE on), I had to lock a few select * from events tests down to just one variant.

We're changing PoE to v2 soonish anyway, and the "custom sql fields" PR will change the way fields are resolved as well, so I'm not too bothered by this right now.

[Twixes]

This seems straightforward

[Twixes]

Maybe one question is: should the raw versions be exposed to users?

[mariusandra]

Maybe one question is: should the raw versions be exposed to users?

Good question. They can be sometimes useful to users, so hard to say. As tables they should just work when selecting (we need to select from somewhere), but we could hide them in the database scene. They could be useful for more advanced users to debug issues though 🤔.

Edit: they could also be useful for "breaking out" of our argMax jail, e.g. to get a speed boost somewhere.