fix(client): close the socket when the browser dies before the close handshake

[route]

WebSocket#close only sends a close frame. The socket itself is closed by on_close, which runs when the browser answers with a close frame of its own — and a browser that is already dead never answers. So #quit and #restart left the socket open until GC finalized the TCPSocket: one leaked descriptor per call, silently, with nothing raised.

That is enough to raise Errno::EMFILE from Process.spawn when browsers are restarted in a loop under a low nofile limit, which is what #637 hit.

Fix is to close the socket explicitly at the end of Client#close, after the grace period the dispatch thread already gets. A healthy quit still completes the handshake through on_close inside that window, so the forced close is a no-op there; it only does work when the peer is gone.

Before

healthy #restart x5:                    fds flat at 8
SIGKILL the browser tree, #restart x6:  9, 10, 11, 12, 13, 14

After

SIGKILL the browser tree, #restart x6:  8, 8, 8, 8, 8, 8

The new spec kills the browser's process group, quits, and asserts the socket is closed. It fails on main with expected #<TCPSocket:fd 13, AF_INET, 127.0.0.1, 60322>.closed? to be truthy, got false.

Refs #637