Check EINTR is handled correctly

[talex5]

System calls need to be retried if they return EINTR (unless this was due to them being cancelled; you don't always get ECANCELED in that case).

There is some handling already, e.g. here's eio_linux handling EINTR for rw ops:

eio/lib_eio_linux/eio_linux.ml

Lines 519 to 527 in 85841dc

	begin match Fiber_context.get_error action.fiber with
	| Some e -> Suspended.discontinue action e (* If cancelled, report that instead. *)
	| None ->
	if errno_is_retry e then (
	submit_rw_req st req;
	schedule st
	) else (
	Suspended.continue action e
	)

Check if we need this in more places, and also check the eio_luv backend.

@hannesm notes in mirage/mirage-crypto#155 (comment) that getrandom needs this too.

[haesbaert]

We could also just enforce SA_RESTART.
Not sure how you feel about eio calling sigaction though.

[talex5]

Reading signal(7), it seems hard to avoid EINTR in all cases. In particular, it seems that SIGCONT can generate EINTR.

It's also a bit unclear to me what happens with signals and io_uring operations. I would expect that the io_uring_enter call would return EINTR rather than the individual operations, so maybe the retry logic above isn't even needed. That code has been there since the very first commit by @avsm; I don't know if it was needed for something.

[talex5]

As far as I know this is working correctly everywhere, so I'll close this. We can reopen if we find any remaining places we missed.