eio_windows: fix anonymous pipes on Windows

[avsm]

This is a testcase failing on my machine atm, so opening to check windows CI here as well (I'm struggling with Windows Redirection Guard nuking my ssh setup due to the use of symlinks in Flexlink!)

[avsm]

Confirming it fails in CI too

 [FAIL]        pipe             0   transfer.
  
    ... with 6 subsequent tests skipped.
  
  ┌──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┐
  │ [FAIL]        pipe             0   transfer.                                                                         │
  └──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘
  [exception] Eio.Io _
              Raised at Stdlib__Effect.Deep.discontinue.(fun) in file "effect.ml", line 63, characters 41-48
              Called from Eio_unix__Private.pipe in file "lib_eio/unix/private.ml", line 14, characters 13-37
              Called from Dune__exe__Test_pipe.test_transfer.(fun) in file "lib_eio_windows/test/test_pipe.ml", line 12, characters 13-29
              Called from Eio__core__Switch.run_internal in file "lib_eio/core/switch.ml", line 133, characters 8-12
              Re-raised at Eio__core__Switch.maybe_raise_exs in file "lib_eio/core/switch.ml", line 119, characters 21-56
              Called from Eio__core__Switch.run_internal in file "lib_eio/core/switch.ml", line 149, characters 4-21
              Called from Eio__core__Cancel.with_cc in file "lib_eio/core/cancel.ml", line 118, characters 8-12
              Re-raised at Eio__core__Cancel.with_cc in file "lib_eio/core/cancel.ml", line 120, characters 50-58
              Called from Alcotest_engine__Core.Make.protect_test.(fun) in file "src/alcotest-engine/core.ml", line 186, characters 17-23
              Called from Alcotest_engine__Monad.Identity.catch in file "src/alcotest-engine/monad.ml", line 24, characters 31-35

[avsm]

I'm going to layer the process spawnign PR over this one; I can't do a full test case for a failing pipe without another process!

[talex5]

This doesn't look right - how will cancellation work?

According to https://stackoverflow.com/questions/34504970/non-blocking-read-on-os-pipe-on-windows, Windows does support non-blocking reads on pipes:

SetNamedPipeHandleState doc allows to put anonymous pipe to non-blocking mode.

Also, we probably want the same check for blocking mode that's in eio_posix. There, we have e.g.

eio/lib_eio_posix/low_level.ml

Lines 67 to 70 in 318abba

	let read fd buf start len =
	if Fd.is_blocking fd then await_readable "read" fd;
	Fd.use_exn "read" fd @@ fun fd ->
	do_nonblocking Read "read" (fun fd -> Unix.read fd buf start len) fd

But in eio_windows, the wait is unconditional:

eio/lib_eio_windows/low_level.ml

Lines 41 to 44 in 318abba

	let read fd buf start len =
	await_readable fd;
	Fd.use_exn "read" fd @@ fun fd ->
	do_nonblocking Read (fun fd -> Unix.read fd buf start len) fd

[talex5]

We do have Eio.Flow.read_all.

[avsm]

Need this expanded to put debug printfs in! Not actually tried cancellations in, will do next: I think I've found one potential weirdness which is that the pipe behaviour changes if I SSH in vs run at the machine directly! Windows is driving me mad, mad I tell you.