waitUntilExit: ignore RunLoop.run()'s return value

[fischman-bcny]

We've been seeing frequent (but not reproducible in isolation) failures in our builds where waitUntilExit returns but terminationStatus fails its precondition on hasFinished because isRunning is still true. This is a speculative fix, since I haven't been able to reproduce the failure in a self-contained test.

This should be safe since Process.waitUntilExit isn't spec'd to relate to RunLoop. In particular there is no guarantee that "stopping" a RunLoop should cause waitUntilExit to return before the monitored process has exited.

---

Unrelated to the above, the second commit in this PR is required to unbreak XCTest's tests, which are run as part of this repo's CI (and so is required to be able to get a green run for this PR).
Absent that commit, XCTest's tests exit -5 on my M1 Pro and -4 on CI (with no output). On my machine, Console reveals:

BUG IN CLIENT OF LIBPTHREAD: workgroup functions already installed
Abort Cause 8313026928

and running under lldb blames CFRuntime.c:1185's call to libdispatch_init().
Breaking on libdispatch_init in lldb confirms libdispatch_init() is called twice, first by swift-corelibs-libdispatch/src/init.c in an __attribute__((constructor)) and the second in the site being disabled in this commit.

[compnerd]

@swift-ci please test

[fischman-bcny]

The reason I believe this PR might help the observed issue is that it's possible to make RunLoop.run(.default) return false with sufficiently esoteric setup, such as:

import Foundation

let waiter = DispatchGroup()
waiter.enter()
let t = Thread {
    defer { waiter.leave() }
    let runLoop = RunLoop.current
    let customMode = RunLoop.Mode("yo")
    runLoop.add(Timer(timeInterval: 0.1, repeats: false) { _ in
        let innerRunResult = runLoop.run(mode: .default, before: Date(timeIntervalSinceNow: 0.1))
        print("inner runLoop.run(.default) returned \(innerRunResult)")
    }, forMode: customMode)

    let outerRunResult = runLoop.run(mode: customMode, before: Date(timeIntervalSinceNow: 1))
    print("outer runLoop.run(custom) returned: \(outerRunResult)")
}
t.start()
waiter.wait()

which emits:

inner runLoop.run(.default) returned false
outer runLoop.run(custom) returned: true

That inner runLoop.run(.default) returned false is what I suspect is similar to what's happening inside Process.waitUntilExit's call to .run(mode: .default, ...) and triggering the too-early return.

[compnerd]

CC: @parkera @iCharlesHu @pcbeard

[compnerd]

@swift-ci please test macOS platform

[compnerd]

@swift-ci please test

[compnerd]

@swift-ci please test Windows platform

[compnerd]

Ideally we would use the deployment version - but this is still forwards/backwards incompatible. @parkera - any thoughts? I'm tempted to say we merge this even with that limitation as this is blocking PRs for other platforms.

[parkera]

This code is designed to go with the corresponding libdispatch, so if we are making a change to both (or a change to dispatch has already been made), they would go together into a Swift release.

[parkera]

LGTM, thanks @fischman-bcny and @compnerd