IO#close() seems entirely wrong here, and very likely is the cause for
our defunct subprocesses; I thought IO#close() did an implicit
Process.wait(), but it seems to actually just flush buffers and
forcefully close immediately, even if the process has not exited yet.

If this commit fixes our issues I'll do a pass through our other uses
of IO#close() for subprocesses and fix them too.

Will-fix: #17551