[Runtime] Resume to the generic executor in the absence of an override.

[varungandhi-apple]

Without this fix, we saw a crash involving URLSession where we'd hit
assertions or corruption in the runtime in various forms:

• Assertion on trying to enqueue work onto a zombie actor.
• job->SchedulerPrivate being overwritten (crash in getNextJobInQueue).
• *(job->SchedulerPrivate) being overwritten (crash in getAsPreprocessedJob
  right after call to getNextJobInQueue).

I haven't been able to create a minimal crashing example yet, but manual
testing shows this does fix the issue.

Fixes rdar://79859254.

[varungandhi-apple]

@swift-ci smoke test

[varungandhi-apple]

Code snippet which I tried to use:

// RUN: %target-swift-frontend -target x86_64-apple-macos12.0 -emit-sil -enable-experimental-concurrency -parse-as-library %s > ~/tmp/tmp.sil
// RUN: %target-build-swift -Xfrontend -enable-experimental-concurrency -Xfrontend -parse-as-library -Xfrontend -disable-availability-checking %s -o %t_binary
// RUN: %{python} %S/../Inputs/timeout.py 2.0 %target-run %t_binary | tee ~/tmp/tmp.stdout | %FileCheck %s
// RUN: DYLD_PRINT_LIBRARIES=1 %target-run %t_binary | tee ~/tmp/tmp.stdout

// REQUIRES: concurrency
// REQUIRES: executable_test
// REQUIRES: OS=macosx || OS=ios

// rdar://76038845
// UNSUPPORTED: use_os_stdlib
// UNSUPPORTED: back_deployment_runtime

// CHECK: Returning from withCheckedContinuation.
// CHECK: Actor deinit.
// CHECK: Returned from A.f().

typealias Cont = CheckedContinuation<Void, Never>

actor A {
    public var cont: Cont? = nil
    init() {}
    func f() async {
        await withCheckedContinuation { (c: Cont) -> Void in
          print("Returning from withCheckedContinuation.")
          self.cont = c
        }
    }
    deinit {
        print("Actor deinit.")
    }
}

@main
struct M {
    static func main() async {
        let cont: Cont?
        do {
          let a = A()
          Task {
            await a.f()
            print("Returned from A.f().")
          }
          let t: Task<Cont?, Never> = Task.detached { 
              try? await Task.sleep(nanoseconds: 10_000_000)
              return await a.cont
          }
          cont = await t.value
        }
        cont!.resume()
        try? await Task.sleep(nanoseconds: 10_000_000)
    }
}

I was hoping that the actor would be de-inited before the continuation is resumed, leading to the continuation trying to resume on the zombie/deinited actor, leading to a crash in the absence of this fix. However, there is a retain/release pair in func f(), and the release only happens after withCheckedContinuation is done. So the actor ends up being de-inited last, and everything works fine, instead of crashing before this change.

As an alternative approach, I considered generating the SIL and tweaking it to do what I wanted. To first establish a baseline, I tried compiling the SIL for the example above to executable code and then running it. However, that segfaults for me instead of printing various strings (i.e. Swift -> executable works OK, SIL -> executable doesn't). I have not had time to look at why that's the case (since landing this patch is somewhat pressing), but that reduced my confidence in being able to write some SIL code and having that compile correctly all the way to executable code.