Fixes setting IO blocking on OpenBSD

[wmoxam]

Fixes #4997

It's worth noting that this issue is what prevents the crystal deps command from working on OpenBSD

[RX14]

Why should this be necessary? Surely this breaks the principle of least surprise. I'd love to hear from someone working on openbsd why this doesn't work.

[jhass]

Do you have any insight as to why calling fcntl(F_SETFL, fcntl(F_GETFL)) is problematic?

At least there should be a comment explaining that is necessary for OpenBSD support, ideally also stating why.

[RX14]

Perhaps try to reproduce in C and ask the mailinglist.

[wmoxam]

If I remember correctly this is only problematic in only one particular state, so perhaps my fix is too broad.

I'll add some test cases and produce a more concise fix

[wmoxam]

For reference, Ruby made a similar change because F_SETFL is an expensive operation since it needs to affect all processes with the same file object

https://bugs.ruby-lang.org/issues/4528

[ysbaddaden]

Looks good to me. Let's avoid a useless syscall when O_NONBLOCK is already set/unset.

[RX14]

@ysbaddaden this is a fine performance optimization, but it shouldn't be necessary to run the program.

[RX14]

It would be great to get some more investigation, but this is fine to merge.