child_process: emit close on parent stdin when child closes fd 0

[maruthang]

Since PR #18701 the parent's child.stdin Socket stopped emitting
close because the Socket is constructed with readable: false,
which suppresses the constructor's read(0) call. Without the libuv
read loop, the kernel's EOF/HUP signal (raised when the child
closes its end of the stdin pipe) is never observed, so the close
cascade never runs.

Install a minimal onread callback on the underlying Pipe handle
and call pipe.readStart() so EOF is delivered. The callback only
acts on UV_EOF and calls socket.destroySoon() so pending writes
drain. Unexpected inbound data is absorbed via readStop() so the
read side-channel is not resurrected.

Fixes: #25131

---

Note: I was unable to run the test suite locally on Linux/macOS (no built out/Release/node on this Windows host; named-pipe EOF semantics on Windows differ, hence the test guard). The fix matches the reporter's documented workaround and Tseian's analysis on the issue thread. Looking forward to CI verification on Linux/macOS runners.

[addaleax]

Still think the approach discussed in #62353 is the way to go here

[maruthang]

Closing in favor of #62353 per @addaleax — the C++ pipe-wrap peer-close approach taken there is the right shape, rather than driving EOF detection from JS via pipe.readStart() + onread. Thanks for the steer.