Write at 0x00c000368528 by goroutine 25:
      os/exec.(*Cmd).Wait()
          os/exec/exec.go:926 +0x1a4
      github.com/onsi/ginkgo/v2/ginkgo/internal.runParallel.func2()
          github.com/onsi/ginkgo/[email protected]/ginkgo/internal/run.go:231 +0x37

    Previous read at 0x00c000368528 by goroutine 96714:
      github.com/onsi/ginkgo/v2/ginkgo/internal.runParallel.func1()
          github.com/onsi/ginkgo/[email protected]/ginkgo/internal/run.go:228 +0x3b
      github.com/onsi/ginkgo/v2/internal/parallel_support.(*ServerHandler).procIsAlive()
          github.com/onsi/ginkgo/[email protected]/internal/parallel_support/server_handler.go:133 +0x121

The problem is that cmd.ProcessState is not thread-safe. The solution is an
atomic bool which gets checked instead and which is written once when the
goroutine waiting for the command has determined that the command has exited.

When a worker unexpectedly dies, it's hard to determine why. Even if output is
available (--output-interceptor-mode=none), if it gets killed by the Linux OOM
killer there's nothing in the process output about that.

Including a textual description of the exit status provides that
information (here simulated with `killall -KILL e2e.test`):

    Output from proc 1:
        I1204 10:03:40.282670  196272 e2e.go:109] Starting e2e run "82689063-1787-40c3-be7d-c4677c556063" on Ginkgo node 1

    Exit result of proc 1:
      signal: killed

In the following tree, the code cleaning up after the inner container node set
a nil current construction node report, causing
CurrentTreeConstructionNodeReport to panic with "CurrentConstructionNodeReport
may only be called during construction of the spec tree" when called by the
transformer for the following It:

    var _ = Describe("outer", func() {
        Context("inner", func() {
            ...
        })
        It("works", func() { ... })
    })

The fix is to restore the old value instead of setting to nil.

…uses precious memory

Signed-off-by: Icarus Wu <[email protected]>