Cats effect 3

[cornerman]

This is the best I could come up with for cats-effect 3 without totally rewriting the whole Observable interface.

I think, a potentially cleaner approach would be to move the effect type into the Observable trait like Observable[F[_], A]. But that has other disadvantages - it becomes difficult to mix different effect types and it adds noise by introducing a type parameter everywhere.

Changes

Functions like fromAsync[F[_]], mapAsync[F[_]], etc. do not just need an Effect[F] instance, but actually need a Dispatcher[F] or IORuntime now. Therefore, we have invented our own replacement unlawful typeclass for effect: RunEffect and RunSyncEffect (this is similar to what scalajs-react does). Coming with instances for IO and SyncIO. You can also create one for Async or with a Dispatcher. Also adding a module for zio Task support.

• fromEffect[F[_]], mapEffect[F[_]] which takes a RunEffectF] implicitly

Thereby fromSync[F[_]] and mapSync[F[_]] are also deprecated in favor of fromEffect/mapEffect.

Furthermore, we have renamed all side-effecting functions like subscribe, onNext, onError, cancel to unsafeSubscribe, unsafeOnNext, unsafeOnError and unsafeCancel. I know, the names are not that beautiful, but I want to make sure that users know that these method actually are a side-effect. The old ones are now deprecated. There also now exist referential transparent versions of these: subscribeF[F[_]:Sync], subscribeIO, subscribeSyncIO and so on.

We had to drop support for monix for now, because it does not support cats-effect 3 yet. I have commented it out for now.

I also took the liberty to remove the types CreateSubject, CreateProSubject. The only use-case was with monix in outwatch and will now be removed.

Feedback is welcome! @TobiasRoland @fdietze What do you think?

Fixes #165
Fixes #5

[cornerman]

Realistically, people will probably mostly use fromIO / mapIO variants.

The methods returning resources are the "correct" implementation for the current interface but feel unergonomic.

The methods taking an implicit dispatcher is a pattern i saw in https://github.com/zio/interop-cats. It's just that you normally do not have a dispatcher implicitly - cats effect actively advises against it.

Possibly makes sense to have extension methods for specific libraries like zio or monix to make it work better.

[cornerman]

Another question: should we just use IORuntime.global as we do now (equivalent of what we did in cats effect 2). Or should we pass it in implicitly?

[cornerman]

If you have an IORuntime in scope now, it will be picked up. Not the most reliable way, but the best we can do for now without requiring one.