Initial windows scheduler

[patricoferris]

As suggested in #125, this is just the mock backend scheduler placed into the Windows IOCP backend so we can try to do this incrementally. I'm seeing if I can get a reliable (if horribly hacked together) Github CI setup for Windows and OCaml 5 too.

[talex5]

Seems like a reasonable place to start, especially if it lets us get Windows CI running.

It might be better to base sched.ml on a simplified version of the Eio_linux one, though (e.g. with Uring.wait replaced with a failwith "TODO"). Then we'd already have the structure for polling for IO, doing wakeups, etc. (assuming uring is somewhat similar to iocp)

[talex5]

After some offline discussion with @avsm and @patricoferris, we decided the best way forward here is to make a simple non-IOCP Windows backend now for Eio 0.10, and then work on IOCP support later. The plan for the initial support is to copy eio_posix as eio_windows and remove anything that doesn't compile. Then anyone with some free time can work on the resulting TODOs.

[avsm]

Agreed. My concern is that the IOCP bindings still need some work, and I think a slower Windows backend is fine for the eio 1.0, followed by an IOCP version in a subsequent release.

[patricoferris]

I have this nearly working and will open a new PR (and close this one once I have), just some uncompilable bits being brought in from eio.unix where I need some of the features (e.g. FD stuff) but not others (e.g. fork actions)

[talex5]

Ah. I guess most of fork_action.c needs wrapping in a #ifdef then.

[patricoferris]

New PR over at #497