Type system

[aljazerzen]

Up until now, PRQL hasn't defined what "relation" is, at least on the language
level. We've been talking about numbers, arithmetic operations and datetimes,
without really defining them and offloading their definitions to what ever
database each of us has in mind. But their definitions are important, with
the simplest example of expression 4 - 5, which has different results depending
on whether integers are signed or not.

So goal of this issue is to settle the definition of what our functions are
operating on. This will allow us to define what results of expressions should
be and then try to implement that behavior for each of the DBMSs.

This will obviously (and hopefully) trigger lots of responses.
To structure the discussion better, I've written something resembling a RFC in PR #1964.
I suggest we use review comments to stay on topic.

This issue can remain open as a tracking issue for new type-system-related issues
and for PRs implementing it.

Ref #381